diff --git a/crudkit/__init__.py b/crudkit/__init__.py
index cf333de..e6af0be 100644
--- a/crudkit/__init__.py
+++ b/crudkit/__init__.py
@@ -1,8 +1,17 @@
-from .mixins import CrudMixin
-from .dsl import QuerySpec
-from .eager import default_eager_policy
-from .service import CrudService
-from .serialize import serialize
-from .blueprint import make_blueprint
+from .backend import BackendInfo, make_backend_info
+from .config import Config, DevConfig, TestConfig, ProdConfig, get_config, build_database_url
+from .engines import CRUDKitRuntime, build_engine, build_sessionmaker
+from .integration import CRUDKit
 
-__all__ = ["CrudMixin", "QuerySpec", "default_eager_policy", "CrudService", "serialize", "make_blueprint"]
\ No newline at end of file
+__all__ = [
+    "Config", "DevConfig", "TestConfig", "ProdConfig", "get_config", "build_database_url",
+    "CRUDKitRuntime", "build_engine", "build_sessionmaker", "BackendInfo", "make_backend_info"
+]
+
+runtime = CRUDKitRuntime()
+crud: CRUDKit | None = None
+
+def init_crud(app):
+    global crud
+    crud = CRUDKit(app, runtime)
+    return crud
diff --git a/crudkit/_sqlite.py b/crudkit/_sqlite.py
new file mode 100644
index 0000000..159156f
--- /dev/null
+++ b/crudkit/_sqlite.py
@@ -0,0 +1,14 @@
+from __future__ import annotations
+from sqlalchemy import event
+from sqlalchemy.engine import Engine
+
+def apply_sqlite_pragmas(engine: Engine, pragmas: dict[str, str]) -> None:
+    if not str(engine.url).startswith("sqlite://"):
+        return
+
+    @event.listens_for(engine, "connect")
+    def set_sqlite_pragma(dbapi_connection, connection_record):
+        cursor = dbapi_connection.cursor()
+        for key, value in pragmas.items():
+            cursor.execute(f"PRAGMA {key}={value}")
+        cursor.close()
diff --git a/crudkit/api/__init__.py b/crudkit/api/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/crudkit/api/_cursor.py b/crudkit/api/_cursor.py
new file mode 100644
index 0000000..3b63cfd
--- /dev/null
+++ b/crudkit/api/_cursor.py
@@ -0,0 +1,21 @@
+import base64, json
+from typing import Any
+
+def encode_cursor(values: list[Any] | None, desc_flags: list[bool], backward: bool) -> str | None:
+    if not values:
+        return None
+    payload = {"v": values, "d": desc_flags, "b": backward}
+    return base64.urlsafe_b64encode(json.dumps(payload).encode()).decode()
+
+def decode_cursor(token: str | None) -> tuple[list[Any] | None, bool] | tuple[None, bool]:
+    if not token:
+        return None, False
+    try:
+        obj = json.loads(base64.urlsafe_b64decode(token.encode()).decode())
+        vals = obj.get("v")
+        backward = bool(obj.get("b", False))
+        if isinstance(vals, list):
+            return vals, backward
+    except Exception:
+        pass
+    return None, False
diff --git a/crudkit/api/flask_api.py b/crudkit/api/flask_api.py
new file mode 100644
index 0000000..bf505b2
--- /dev/null
+++ b/crudkit/api/flask_api.py
@@ -0,0 +1,90 @@
+from flask import Blueprint, jsonify, request
+
+from crudkit.api._cursor import encode_cursor, decode_cursor
+from crudkit.core.service import _is_truthy
+
+def generate_crud_blueprint(model, service):
+    bp = Blueprint(model.__name__.lower(), __name__)
+
+    @bp.get('/')
+    def list_items():
+        args = request.args.to_dict(flat=True)
+
+        # legacy detection
+        legacy_offset = "offset" in args or "page" in args
+
+        # sane limit default
+        try:
+            limit = int(args.get("limit", 50))
+        except Exception:
+            limit = 50
+        args["limit"] = limit
+
+        if legacy_offset:
+            # Old behavior: honor limit/offset, same CRUDSpec goodies
+            items = service.list(args)
+            return jsonify([obj.as_dict() for obj in items])
+
+        # New behavior: keyset seek with cursors
+        key, backward = decode_cursor(args.get("cursor"))
+
+        window = service.seek_window(
+            args,
+            key=key,
+            backward=backward,
+            include_total=_is_truthy(args.get("include_total", "1")),
+        )
+
+        desc_flags = list(window.order.desc)
+        body = {
+            "items": [obj.as_dict() for obj in window.items],
+            "limit": window.limit,
+            "next_cursor": encode_cursor(window.last_key, desc_flags, backward=False),
+            "prev_cursor": encode_cursor(window.first_key, desc_flags, backward=True),
+            "total": window.total,
+        }
+
+        resp = jsonify(body)
+        # Optional Link header
+        links = []
+        if body["next_cursor"]:
+            links.append(f'<{request.base_url}?cursor={body["next_cursor"]}&limit={window.limit}>; rel="next"')
+        if body["prev_cursor"]:
+            links.append(f'<{request.base_url}?cursor={body["prev_cursor"]}&limit={window.limit}>; rel="prev"')
+        if links:
+            resp.headers["Link"] = ", ".join(links)
+        return resp
+
+    @bp.get('/<int:id>')
+    def get_item(id):
+        item = service.get(id, request.args)
+        try:
+            return jsonify(item.as_dict())
+        except Exception as e:
+            return jsonify({"status": "error", "error": str(e)})
+
+    @bp.post('/')
+    def create_item():
+        obj = service.create(request.json)
+        try:
+            return jsonify(obj.as_dict())
+        except Exception as e:
+            return jsonify({"status": "error", "error": str(e)})
+
+    @bp.patch('/<int:id>')
+    def update_item(id):
+        obj = service.update(id, request.json)
+        try:
+            return jsonify(obj.as_dict())
+        except Exception as e:
+            return jsonify({"status": "error", "error": str(e)})
+
+    @bp.delete('/<int:id>')
+    def delete_item(id):
+        service.delete(id)
+        try:
+            return jsonify({"status": "success"}), 204
+        except Exception as e:
+            return jsonify({"status": "error", "error": str(e)})
+
+    return bp
diff --git a/crudkit/backend.py b/crudkit/backend.py
new file mode 100644
index 0000000..3232b71
--- /dev/null
+++ b/crudkit/backend.py
@@ -0,0 +1,122 @@
+from __future__ import annotations
+from dataclasses import dataclass
+from typing import Tuple, Optional, Iterable
+from contextlib import contextmanager
+from sqlalchemy import text, func
+from sqlalchemy.engine import Engine
+from sqlalchemy.orm import Session
+from sqlalchemy.sql.elements import ClauseElement
+from sqlalchemy.sql import Select
+
+@dataclass(frozen=True)
+class BackendInfo:
+    name: str
+    version: Tuple[int, ...]
+    paramstyle: str
+    is_sqlite: bool
+    is_postgres: bool
+    is_mysql: bool
+    is_mssql: bool
+
+    supports_returning: bool
+    supports_ilike: bool
+    requires_order_by_for_offset: bool
+    max_bind_params: Optional[int]
+
+    @classmethod
+    def from_engine(cls, engine: Engine) -> "BackendInfo":
+        d = engine.dialect
+        name = d.name
+        version = tuple(getattr(d, "server_version_info", ()) or ())
+        is_pg = name in {"postgresql", "postgres"}
+        is_my = name == "mysql"
+        is_sq = name == "sqlite"
+        is_ms = name == "mssql"
+
+        supports_ilike = is_pg or is_my
+        supports_returning = is_pg or (is_sq and version >= (3, 35))
+        requires_order_by_for_offset = is_ms
+
+        max_bind_params = 999 if is_sq else None
+
+        return cls(
+            name=name,
+            version=version,
+            paramstyle=d.paramstyle,
+            is_sqlite=is_sq,
+            is_postgres=is_pg,
+            is_mysql=is_my,
+            is_mssql=is_ms,
+            supports_returning=supports_returning,
+            supports_ilike=supports_ilike,
+            requires_order_by_for_offset=requires_order_by_for_offset,
+            max_bind_params=max_bind_params,
+        )
+
+def make_backend_info(engine: Engine) -> BackendInfo:
+    return BackendInfo.from_engine(engine)
+
+def ci_like(column, value: str, backend: BackendInfo) -> ClauseElement:
+    """
+    Portable save-insensitive LIKE.
+    Uses ILIKE where available, else lower() dance.
+    """
+    pattern = f"%{value}%"
+    if backend.supports_ilike:
+        return column.ilike(pattern)
+    return func.lower(column).like(func.lower(text(":pattern"))).params(pattern=pattern)
+
+def apply_pagination(sel: Select, backend: BackendInfo, *, page: int, per_page: int, default_order_by=None) -> Select:
+    """
+    Portable pagination. MSSQL requires ORDER BY when using OFFSET
+    """
+    page = max(1, int(page))
+    per_page = max(1, int(per_page))
+    offset = (page - 1) * per_page
+
+    if backend.requires_order_by_for_offset and not sel._order_by_clauses:
+        if default_order_by is None:
+            sel = sel.order_by(text("1"))
+        else:
+            sel = sel.order_by(default_order_by)
+
+    return sel.limit(per_page).offset(offset)
+
+@contextmanager
+def maybe_identify_insert(session: Session, table, backend: BackendInfo):
+    """
+    For MSSQL tables with IDENTIFY PK when you need to insert explicit IDs.
+    No-op elsewhere.
+    """
+    if not backend.is_mssql:
+        yield
+        return
+
+    full_name = f"{table.schema}.{table.name}" if table.schema else table.name
+    session.execute(text(f"SET IDENTIFY_INSERT {full_name} ON"))
+    try:
+        yield
+    finally:
+        session.execute(text(f"SET IDENTITY_INSERT {full_name} OFF"))
+
+def chunked_in(column, values: Iterable, backend: BackendInfo, chunk_size: Optional[int] = None) -> ClauseElement:
+    """
+    Build a safe large IN() filter respecting bund param limits.
+    Returns a disjunction of chunked IN clauses if needed.
+    """
+    vals = list(values)
+    if not vals:
+        return text("1=0")
+
+    limit = chunk_size or backend.max_bind_params or len(vals)
+    if len(vals) <= limit:
+        return column.in_(vals)
+
+    parts = []
+    for i in range(0, len(vals), limit):
+        parts.append(column.in_(vals[i:i + limit]))
+
+    expr = parts[0]
+    for p in parts[1:]:
+        expr = expr | p
+    return expr
diff --git a/crudkit/blueprint.py b/crudkit/blueprint.py
deleted file mode 100644
index cae482e..0000000
--- a/crudkit/blueprint.py
+++ /dev/null
@@ -1,81 +0,0 @@
-from flask import Blueprint, request, jsonify, abort
-from sqlalchemy.orm import scoped_session
-from .dsl import QuerySpec
-from .service import CrudService
-from .eager import default_eager_policy
-from .serialize import serialize
-
-def make_blueprint(db_session_factory, registry):
-    bp = Blueprint("crud", __name__)
-    def session(): return scoped_session(db_session_factory)()
-
-    @bp.get("/<model>/list")
-    def list_items(model):
-        Model = registry.get(model) or abort(404)
-        spec = QuerySpec(
-            filters=_parse_filters(request.args),
-            order_by=request.args.getlist("sort"),
-            page=request.args.get("page", type=int),
-            per_page=request.args.get("per_page", type=int),
-            expand=request.args.getlist("expand"),
-            fields=request.args.get("fields", type=lambda s: [x.strip() for x in s.split(",")] if s else None),
-        )
-        s = session(); svc = CrudService(s, default_eager_policy)
-        rows, total = svc.list(Model, spec)
-        data = [serialize(r, fields=spec.fields, expand=spec.expand) for r in rows]
-        return jsonify({"data": data, "total": total})
-
-    @bp.post("/<model>")
-    def create_item(model):
-        Model = registry.get(model) or abort(404)
-        payload = request.get_json() or {}
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.create(Model, payload)
-        s.commit()
-        return jsonify(serialize(obj)), 201
-
-    @bp.get("/<model>/<int:id>")
-    def read_item(model, id):
-        Model = registry.get(model) or abort(404)
-        spec = QuerySpec(expand=request.args.getlist("expand"),
-                         fields=request.args.get("fields", type=lambda s: s.split(",")))
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.get(Model, id, spec) or abort(404)
-        return jsonify(serialize(obj, fields=spec.fields, expand=spec.expand))
-
-    @bp.patch("/<model>/<int:id>")
-    def update_item(model, id):
-        Model = registry.get(model) or abort(404)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.get(Model, id, QuerySpec()) or abort(404)
-        payload = request.get_json() or {}
-        svc.update(obj, payload)
-        s.commit()
-        return jsonify(serialize(obj))
-
-    @bp.delete("/<model>/<int:id>")
-    def delete_item(model, id):
-        Model = registry.get(model) or abort(404)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.get(Model, id, QuerySpec()) or abort(404)
-        svc.soft_delete(obj)
-        s.commit()
-        return jsonify({"status": "deleted"})
-
-    @bp.post("/<model>/<int:id>/undelete")
-    def undelete_item(model, id):
-        Model = registry.get(model) or abort(404)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.get(Model, id, QuerySpec()) or abort(404)
-        svc.undelete(obj)
-        s.commit()
-        return jsonify({"status": "restored"})
-    return bp
-
-def _parse_filters(args):
-    out = {}
-    for k, v in args.items():
-        if k in {"page", "per_page", "sort", "expand", "fields"}:
-            continue
-        out[k] = v
-    return out
\ No newline at end of file
diff --git a/crudkit/config.py b/crudkit/config.py
new file mode 100644
index 0000000..0439a3e
--- /dev/null
+++ b/crudkit/config.py
@@ -0,0 +1,243 @@
+from __future__ import annotations
+import os
+import os
+from typing import Dict, Any, Optional, Type
+from urllib.parse import quote_plus
+from pathlib import Path
+
+try:
+    from dotenv import load_dotenv
+except Exception:
+    load_dotenv = None
+
+def _load_dotenv_if_present() -> None:
+    """
+    Load .env once if present. Priority rules:
+    1) CRUDKIT_DOTENV points to a file
+    2) Project root's .env (two dirs up from this file)
+    3) Current working directory .env
+
+    Env already present in the process takes precedence.
+    """
+    if load_dotenv is None:
+        return
+
+    path_hint = os.getenv("CRUDKIT_DOTENV")
+    if path_hint:
+        p = Path(path_hint).resolve()
+        if p.exists():
+            load_dotenv(dotenv_path=p, override=True)
+            os.environ["CRUDKIT_DOTENV_LOADED"] = str(p)
+            return
+
+    repo_env = Path(__file__).resolve().parents[1] / ".env"
+    if repo_env.exists():
+        load_dotenv(dotenv_path=repo_env, override=True)
+        os.environ["CRUDKIT_DOTENV_LOADED"] = str(repo_env)
+        return
+
+    cwd_env = Path.cwd() / ".env"
+    if cwd_env.exists():
+        load_dotenv(dotenv_path=cwd_env, override=True)
+        os.environ["CRUDKIT_DOTENV_LOADED"] = str(cwd_env)
+
+def _getenv(name: str, default: Optional[str] = None) -> Optional[str]:
+    """Treat empty strings as missing. Hekos when OS env has DB_BACKEND=''."""
+    val = os.getenv(name)
+    if val is None or val.strip() == "":
+        return default
+    return val
+
+def build_database_url(
+    *,
+    backend: Optional[str] = None,
+    url: Optional[str] = None,
+    user: Optional[str] = None,
+    password: Optional[str] = None,
+    host: Optional[str] = None,
+    port: Optional[str] = None,
+    database: Optional[str] = None,
+    driver: Optional[str] = None,
+    dsn: Optional[str] = None,
+    trusted: Optional[bool] = None,
+    options: Optional[Dict[str, str]] = None,
+) -> str:
+    """
+    Build a SQLAlchemy URL string. If "url" is provided, it wins.
+    Supported: sqlite, postgresql, mysql, mssql (pyodbc)
+    """
+
+    if url:
+        return url
+
+    backend = (backend or "").lower().strip()
+    options = options or {}
+
+    if backend == "sqlite":
+        db_path = database or "app.db"
+        if db_path == ":memory:":
+            return "sqlite:///:memory:"
+        return f"sqlite:///{db_path}"
+
+    if backend in {"postgres", "postgresql"}:
+        driver = driver or "psycopg"
+        user = user or ""
+        password = password or ""
+        creds = f"{quote_plus(user)}:{quote_plus(password)}@" if user or password else ""
+        host = host or "localhost"
+        port = port or "5432"
+        database = database or "app"
+        qs = ""
+        if options:
+            qs = "?" + "&".join(f"{k}={quote_plus(v)}" for k, v in options.items())
+        return f"postgresql+{driver}://{creds}{host}:{port}/{database}{qs}"
+
+    if backend == "mysql":
+        driver = driver or "pymysql"
+        user = user or ""
+        password = password or ""
+        creds = f"{quote_plus(user)}:{quote_plus(password)}@" if user or password else ""
+        host = host or "localhost"
+        port = port or "3306"
+        database = database or "app"
+        qs = ""
+        if options:
+            qs = "?" + "&".join(f"{k}={quote_plus(v)}" for k, v in options.items())
+        return f"mysql+{driver}://{creds}{host}:{port}/{database}{qs}"
+
+    if backend in {"mssql", "sqlserver", "sqlsrv"}:
+        if dsn:
+            qs = ""
+            if options:
+                qs = "?" + "&".join(f"{k}={quote_plus(v)}" for k, v in options.items())
+            return f"mssql+pyodbc://@{quote_plus(dsn)}{qs}"
+
+        driver = driver or "ODBC Driver 18 for SQL Server"
+        host = host or "localhost"
+        port = port or "1433"
+        database = database or "app"
+
+        if trusted:
+            base_opts = {
+                "driver": driver,
+                "Trusted_Connection": "yes",
+                "Encrypt": "yes",
+                "TrustServerCertificate": "yes",
+                "MARS_Connection": "yes",
+            }
+            base_opts.update(options)
+            qs = "?" + "&".join(f"{k}={quote_plus(v)}" for k, v in base_opts.items())
+            return f"mssql+pyodbc://{host}:{port}/{database}{qs}"
+
+        user = user or ""
+        password = password or ""
+        creds = f"{quote_plus(user)}:{quote_plus(password)}@" if user or password else ""
+        base_opts = {
+            "driver": driver,
+            "Encrypt": "yes",
+            "TrustServerCertificate": "yes",
+            "MARS_Connection": "yes",
+        }
+        base_opts.update(options)
+        qs = "?" + "&".join(f"{k}={quote_plus(v)}" for k, v in base_opts.items())
+        return f"mssql+pyodbc://{creds}{host}:{port}/{database}{qs}"
+
+    raise ValueError(f"Unsupported backend: {backend!r}")
+
+class Config:
+    """
+    CRUDKit config: environment-first with sane defaults.
+    Designed to be subclassed by apps, but fine as-is.
+    """
+
+    _dotenv_loaded = False
+
+    DEBUG = False
+    TESTING = False
+    SECRET_KEY = _getenv("SECRET_KEY", "dev-not-secret")
+
+    if not _dotenv_loaded:
+        _load_dotenv_if_present()
+        _dotenv_loaded = True
+
+    DATABASE_URL = build_database_url(
+        url=_getenv("DATABASE_URL"),
+        backend=_getenv("DB_BACKEND"),
+        user=_getenv("DB_USER"),
+        password=_getenv("DB_PASS"),
+        host=_getenv("DB_HOST"),
+        port=_getenv("DB_PORT"),
+        database=_getenv("DB_NAME"),
+        driver=_getenv("DB_DRIVER"),
+        dsn=_getenv("DB_DSN"),
+        trusted=bool(int(_getenv("DB_TRUSTED", "0"))),
+        options=None,
+    )
+
+    SQLALCHEMY_ECHO = bool(int(os.getenv("DB_ECHO", "0")))
+    POOL_SIZE = int(os.getenv("DB_POOL_SIZE", "5"))
+    MAX_OVERFLOW = int(os.getenv("DB_MAX_OVERFLOW", "10"))
+    POOL_TIMEOUT = int(os.getenv("DB_POOL_TIMEOUT", "30"))
+    POOL_RECYCLE = int(os.getenv("DB_POOL_RECYCLE", "1000"))
+    POOL_PRE_PING = True
+
+    SQLITE_PRAGMAS = {
+        "journal_mode": os.getenv("SQLITE_JOURNAL_MODE", "WAL"),
+        "foreign_keys": os.getenv("SQLITE_FOREIGN_KEYS", "ON"),
+        "synchronous": os.getenv("SQLITE_SYNCHRONOUS", "NORMAL"),
+    }
+
+    @classmethod
+    def engine_kwargs(cls) -> Dict[str, Any]:
+        url = cls.DATABASE_URL
+        kwargs: Dict[str, Any] = {
+            "echo": cls.SQLALCHEMY_ECHO,
+            "pool_pre_ping": cls.POOL_PRE_PING,
+            "future": True,
+        }
+
+        if url.startswith("sqlite://"):
+            kwargs["connect_args"] = {"check_same_thread": False}
+
+        kwargs.update(
+            {
+                "pool_size": cls.POOL_SIZE,
+                "max_overflow": cls.MAX_OVERFLOW,
+                "pool_timeout": cls.POOL_TIMEOUT,
+                "pool_recycle": cls.POOL_RECYCLE,
+            }
+        )
+        return kwargs
+
+    @classmethod
+    def session_kwargs(cls) -> Dict[str, Any]:
+        return {
+            "autoflush": False,
+            "autocommit": False,
+            "expire_on_commit": False,
+            "future": True,
+        }
+
+class DevConfig(Config):
+    DEBUG = True
+    SQLALCHEMY_ECHO = bool(int(os.getenv("DB_ECHO", "1")))
+
+class TestConfig(Config):
+    TESTING = True
+    DATABASE_URL = build_database_url(backend="sqlite", database=":memory:")
+    SQLALCHEMY_ECHO = False
+
+class ProdConfig(Config):
+    DEBUG = False
+    SQLALCHEMY_ECHO = bool(int(os.getenv("DB_ECHO", "0")))
+
+def get_config(name: str | None) -> Type[Config]:
+    """
+    Resolve config by name. None -> environment variable CRUDKIT_ENV or 'dev'.
+    """
+    env = (name or os.getenv("CRUDKIT_ENV") or "dev").lower()
+    if env in {"prod", "production"}:
+        return ProdConfig
+    if env in {"test", "testing", "ci"}:
+        return TestConfig
+    return DevConfig
diff --git a/crudkit/core/__init__.py b/crudkit/core/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/crudkit/core/base.py b/crudkit/core/base.py
new file mode 100644
index 0000000..46874fe
--- /dev/null
+++ b/crudkit/core/base.py
@@ -0,0 +1,56 @@
+from sqlalchemy import Column, Integer, DateTime, Boolean, String, JSON, func
+from sqlalchemy.orm import declarative_mixin, declarative_base
+
+Base = declarative_base()
+
+@declarative_mixin
+class CRUDMixin:
+    id = Column(Integer, primary_key=True)
+    created_at = Column(DateTime, default=func.now(), nullable=False)
+    updated_at = Column(DateTime, default=func.now(), nullable=False, onupdate=func.now())
+
+    def as_dict(self, fields: list[str] | None = None):
+        """
+        Serialize the instance.
+            - If 'fields' (possibly dotted) is provided, emit exactly those keys.
+            - Else, if '__crudkit_projection__' is set on the instance, emit those keys.
+            - Else, fall back to all mapped columns on this class hierarchy.
+        Always includes 'id' when present unless explicitly excluded.
+        """
+        if fields is None:
+            fields = getattr(self, "__crudkit_projection__", None)
+
+        if fields:
+            out = {}
+            if "id" not in fields and hasattr(self, "id"):
+                out["id"] = getattr(self, "id")
+            for f in fields:
+                cur = self
+                for part in f.split("."):
+                    if cur is None:
+                        break
+                    cur = getattr(cur, part, None)
+                out[f] = cur
+            return out
+
+        result = {}
+        for cls in self.__class__.__mro__:
+            if hasattr(cls, "__table__"):
+                for column in cls.__table__.columns:
+                    name = column.name
+                    result[name] = getattr(self, name)
+        return result
+
+
+class Version(Base):
+    __tablename__ = "versions"
+
+    id = Column(Integer, primary_key=True)
+    model_name = Column(String, nullable=False)
+    object_id = Column(Integer, nullable=False)
+    change_type = Column(String, nullable=False)
+    data = Column(JSON, nullable=True)
+    timestamp = Column(DateTime, default=func.now())
+
+    actor = Column(String, nullable=True)
+    meta = Column('metadata', JSON, nullable=True)
diff --git a/crudkit/core/metadata.py b/crudkit/core/metadata.py
new file mode 100644
index 0000000..e69de29
diff --git a/crudkit/core/service.py b/crudkit/core/service.py
new file mode 100644
index 0000000..1220e04
--- /dev/null
+++ b/crudkit/core/service.py
@@ -0,0 +1,556 @@
+from typing import Any, Callable, Dict, Iterable, List, Tuple, Type, TypeVar, Generic, Optional, Protocol, runtime_checkable, cast
+from sqlalchemy import and_, func, inspect, or_, text
+from sqlalchemy.engine import Engine, Connection
+from sqlalchemy.orm import Load, Session, raiseload, selectinload, with_polymorphic, Mapper, RelationshipProperty, class_mapper
+from sqlalchemy.orm.attributes import InstrumentedAttribute
+from sqlalchemy.orm.util import AliasedClass
+from sqlalchemy.sql import operators
+from sqlalchemy.sql.elements import UnaryExpression
+
+from crudkit.core.base import Version
+from crudkit.core.spec import CRUDSpec
+from crudkit.core.types import OrderSpec, SeekWindow
+from crudkit.backend import BackendInfo, make_backend_info
+
+def _loader_options_for_fields(root_alias, model_cls, fields: list[str]) -> list[Load]:
+    """
+    For bare MANYTOONE names in fields (e.g. "location"), selectinload the relationship
+    and only fetch the related PK. This is enough for preselecting <select> inputs
+    without projecting the FK column on the root model.
+    """
+    opts: list[Load] = []
+    if not fields:
+        return opts
+
+    mapper = class_mapper(model_cls)
+    for name in fields:
+        prop = mapper.relationships.get(name)
+        if not isinstance(prop, RelationshipProperty):
+            continue
+        if prop.direction.name != "MANYTOONE":
+            continue
+
+        rel_attr = getattr(root_alias, name)
+        target_cls = prop.mapper.class_
+        # load_only PK if present; else just selectinload
+        id_attr = getattr(target_cls, "id", None)
+        if id_attr is not None:
+            opts.append(selectinload(rel_attr).load_only(id_attr))
+        else:
+            opts.append(selectinload(rel_attr))
+
+    return opts
+
+@runtime_checkable
+class _HasID(Protocol):
+    id: int
+
+@runtime_checkable
+class _HasTable(Protocol):
+    __table__: Any
+
+@runtime_checkable
+class _HasADict(Protocol):
+    def as_dict(self) -> dict: ...
+
+@runtime_checkable
+class _SoftDeletable(Protocol):
+    is_deleted: bool
+
+class _CRUDModelProto(_HasID, _HasTable, _HasADict, Protocol):
+    """Minimal surface that our CRUD service relies on. Soft-delete is optional."""
+    pass
+
+T = TypeVar("T", bound=_CRUDModelProto)
+
+def _is_truthy(val):
+    return str(val).lower() in ('1', 'true', 'yes', 'on')
+
+class CRUDService(Generic[T]):
+    def __init__(
+        self,
+        model: Type[T],
+        session_factory: Callable[[], Session],
+        polymorphic: bool = False,
+        *,
+        backend: Optional[BackendInfo] = None
+    ):
+        self.model = model
+        self._session_factory = session_factory
+        self.polymorphic = polymorphic
+        self.supports_soft_delete = hasattr(model, 'is_deleted')
+        # Cache backend info once. If not provided, derive from session bind.
+        bind = self.session.get_bind()
+        eng: Engine = bind.engine if isinstance(bind, Connection) else cast(Engine, bind)
+        self.backend = backend or make_backend_info(eng)
+
+    @property
+    def session(self) -> Session:
+        return self._session_factory()
+
+    def get_query(self):
+        if self.polymorphic:
+            poly = with_polymorphic(self.model, "*")
+            return self.session.query(poly), poly
+        return self.session.query(self.model), self.model
+
+    def _resolve_required_includes(self, root_alias: Any, rel_field_names: Dict[Tuple[str, ...], List[str]]) -> List[Any]:
+        """
+        For each dotted path like ("location"), -> ["label"], look up the target
+        model's __crudkit_field_requires__ for the terminal field and produce
+        selectinload options prefixed with the relationship path, e.g.:
+            Room.__crudkit_field_requires__['label'] = ['room_function']
+            => selectinload(root.location).selectinload(Room.room_function)
+        """
+        opts: List[Any] = []
+
+        root_mapper: Mapper[Any] = cast(Mapper[Any], inspect(self.model))
+
+        for path, names in (rel_field_names or {}).items():
+            if not path:
+                continue
+
+            current_alias = root_alias
+            current_mapper = root_mapper
+            rel_props: List[RelationshipProperty] = []
+
+            valid = True
+            for step in path:
+                rel = current_mapper.relationships.get(step)
+                if rel is None:
+                    valid = False
+                    break
+                rel_props.append(rel)
+                current_mapper = cast(Mapper[Any], inspect(rel.entity.entity))
+            if not valid:
+                continue
+
+            target_cls = current_mapper.class_
+
+            requires = getattr(target_cls, "__crudkit_field_requires__", None)
+            if not isinstance(requires, dict):
+                continue
+
+            for field_name in names:
+                needed: Iterable[str] = requires.get(field_name, [])
+                for rel_need in needed:
+                    loader = selectinload(getattr(root_alias, rel_props[0].key))
+                    for rp in rel_props[1:]:
+                        loader = loader.selectinload(getattr(getattr(root_alias, rp.parent.class_.__name__.lower(), None) or rp.parent.class_, rp.key))
+
+                    loader = loader.selectinload(getattr(target_cls, rel_need))
+                    opts.append(loader)
+
+        return opts
+
+    def _extract_order_spec(self, root_alias, given_order_by):
+        """
+        SQLAlchemy 2.x only:
+        Normalize order_by into (cols, desc_flags). Supports plain columns and
+        col.asc()/col.desc() (UnaryExpression). Avoids boolean evaluation of clauses.
+        """
+        from sqlalchemy.sql import operators
+        from sqlalchemy.sql.elements import UnaryExpression
+
+        given = self._stable_order_by(root_alias, given_order_by)
+
+        cols, desc_flags = [], []
+        for ob in given:
+            # Unwrap column if this is a UnaryExpression produced by .asc()/.desc()
+            elem = getattr(ob, "element", None)
+            col = elem if elem is not None else ob  # don't use "or" with SA expressions
+
+            # Detect direction in SA 2.x
+            is_desc = False
+            dir_attr = getattr(ob, "_direction", None)
+            if dir_attr is not None:
+                is_desc = (dir_attr is operators.desc_op) or (getattr(dir_attr, "name", "").upper() == "DESC")
+            elif isinstance(ob, UnaryExpression):
+                op = getattr(ob, "operator", None)
+                is_desc = (op is operators.desc_op) or (getattr(op, "name", "").upper() == "DESC")
+
+            cols.append(col)
+            desc_flags.append(bool(is_desc))
+
+        from crudkit.core.types import OrderSpec
+        return OrderSpec(cols=tuple(cols), desc=tuple(desc_flags))
+
+    def _key_predicate(self, spec: OrderSpec, key_vals: list[Any], backward: bool):
+        """
+        Build lexicographic predicate for keyset seek.
+        For backward traversal, import comparisons.
+        """
+        if not key_vals:
+            return None
+        conds = []
+        for i, col in enumerate(spec.cols):
+            ties = [spec.cols[j] == key_vals[j] for j in range(i)]
+            is_desc = spec.desc[i]
+            if not backward:
+                op = col < key_vals[i] if is_desc else col > key_vals[i]
+            else:
+                op = col > key_vals[i] if is_desc else col < key_vals[i]
+            conds.append(and_(*ties, op))
+        return or_(*conds)
+
+    def _pluck_key(self, obj: Any, spec: OrderSpec) -> list[Any]:
+        out = []
+        for c in spec.cols:
+            key = getattr(c, "key", None) or getattr(c, "name", None)
+            out.append(getattr(obj, key))
+        return out
+
+    def seek_window(
+        self,
+        params: dict | None = None,
+        *,
+        key: list[Any] | None = None,
+        backward: bool = False,
+        include_total: bool = True,
+    ) -> "SeekWindow[T]":
+        """
+        Transport-agnostic keyset pagination that preserves all the goodies from `list()`:
+          - filters, includes, joins, field projection, eager loading, soft-delete
+          - deterministic ordering (user sort + PK tiebreakers)
+          - forward/backward seek via `key` and `backward`
+        Returns a SeekWindow with items, first/last keys, order spec, limit, and optional total.
+        """
+        session = self.session
+        query, root_alias = self.get_query()
+
+        spec = CRUDSpec(self.model, params or {}, root_alias)
+
+        filters = spec.parse_filters()
+        order_by = spec.parse_sort()
+
+        root_fields, rel_field_names, root_field_names = spec.parse_fields()
+
+        for path, names in (rel_field_names or {}).items():
+            if "label" in names:
+                rel_name = path[0]
+                rel_attr = getattr(root_alias, rel_name, None)
+                if rel_attr is not None:
+                    query = query.options(selectinload(rel_attr))
+
+        # Soft delete filter
+        if self.supports_soft_delete and not _is_truthy(params.get("include_deleted")):
+            query = query.filter(getattr(root_alias, "is_deleted") == False)
+
+        # Parse filters first
+        if filters:
+            query = query.filter(*filters)
+
+        # Includes + joins (so relationship fields like brand.name, location.label work)
+        spec.parse_includes()
+        for parent_alias, relationship_attr, target_alias in spec.get_join_paths():
+            rel_attr = cast(InstrumentedAttribute, relationship_attr)
+            target = cast(Any, target_alias)
+            query = query.join(target, rel_attr.of_type(target), isouter=True)
+
+        # Fields/projection: load_only for root columns, eager loads for relationships
+        only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
+        if only_cols:
+            query = query.options(Load(root_alias).load_only(*only_cols))
+        for eager in spec.get_eager_loads(root_alias, fields_map=rel_field_names):
+            query = query.options(eager)
+
+        for opt in self._resolve_required_includes(root_alias, rel_field_names):
+            query = query.options(opt)
+
+        # Order + limit
+        order_spec = self._extract_order_spec(root_alias, order_by)  # SA2-only helper
+        limit, _ = spec.parse_pagination()
+        if not limit or limit <= 0:
+            limit = 50  # sensible default
+
+        # Keyset predicate
+        if key:
+            pred = self._key_predicate(order_spec, key, backward)
+            if pred is not None:
+                query = query.filter(pred)
+
+        # Apply ordering. For backward, invert SQL order then reverse in-memory for display.
+        if not backward:
+            clauses = []
+            for col, is_desc in zip(order_spec.cols, order_spec.desc):
+                clauses.append(col.desc() if is_desc else col.asc())
+            query = query.order_by(*clauses).limit(limit)
+            items = query.all()
+        else:
+            inv_clauses = []
+            for col, is_desc in zip(order_spec.cols, order_spec.desc):
+                inv_clauses.append(col.asc() if is_desc else col.desc())
+            query = query.order_by(*inv_clauses).limit(limit)
+            items = list(reversed(query.all()))
+
+        # Tag projection so your renderer knows what fields were requested
+        proj = []
+        if root_field_names:
+            proj.extend(root_field_names)
+        if root_fields:
+            proj.extend(c.key for c in root_fields)
+        for path, names in (rel_field_names or {}).items():
+            prefix = ".".join(path)
+            for n in names:
+                proj.append(f"{prefix}.{n}")
+        if proj and "id" not in proj and hasattr(self.model, "id"):
+            proj.insert(0, "id")
+        if proj:
+            for obj in items:
+                try:
+                    setattr(obj, "__crudkit_projection__", tuple(proj))
+                except Exception:
+                    pass
+
+        # Boundary keys for cursor encoding in the API layer
+        first_key = self._pluck_key(items[0], order_spec) if items else None
+        last_key  = self._pluck_key(items[-1], order_spec) if items else None
+
+        # Optional total that’s safe under JOINs (COUNT DISTINCT ids)
+        total = None
+        if include_total:
+            base = self.session.query(getattr(root_alias, "id"))
+            if self.supports_soft_delete and not _is_truthy(params.get("include_deleted")):
+                base = base.filter(getattr(root_alias, "is_deleted") == False)
+            if filters:
+                base = base.filter(*filters)
+            # replicate the same joins used above
+            for parent_alias, relationship_attr, target_alias in spec.get_join_paths():
+                rel_attr = cast(InstrumentedAttribute, relationship_attr)
+                target = cast(Any, target_alias)
+                base = base.join(target, rel_attr.of_type(target), isouter=True)
+            total = self.session.query(func.count()).select_from(base.order_by(None).distinct().subquery()).scalar() or 0
+
+        from crudkit.core.types import SeekWindow  # avoid circulars at module top
+        return SeekWindow(
+            items=items,
+            limit=limit,
+            first_key=first_key,
+            last_key=last_key,
+            order=order_spec,
+            total=total,
+        )
+
+    # Helper: default ORDER BY for MSSQL when paginating without explicit order
+    def _default_order_by(self, root_alias):
+        mapper: Mapper[Any] = cast(Mapper[Any], inspect(self.model))
+        cols = []
+        for col in mapper.primary_key:
+            try:
+                cols.append(getattr(root_alias, col.key))
+            except AttributeError:
+                cols.append(col)
+        return cols or [text("1")]
+
+    def _stable_order_by(self, root_alias, given_order_by):
+        """
+        Ensure deterministic ordering by appending PK columns as tiebreakers.
+        If no order is provided, fall back to default primary-key order.
+        """
+        order_by = list(given_order_by or [])
+        if not order_by:
+            return self._default_order_by(root_alias)
+
+        mapper: Mapper[Any] = cast(Mapper[Any], inspect(self.model))
+        pk_cols = []
+        for col in mapper.primary_key:
+            try:
+                pk_cols.append(getattr(root_alias, col.key))
+            except ArithmeticError:
+                pk_cols.append(col)
+
+        return [*order_by, *pk_cols]
+
+    def get(self, id: int, params=None) -> T | None:
+        query, root_alias = self.get_query()
+
+        include_deleted = False
+        root_fields = []
+        root_field_names = {}
+        rel_field_names = {}
+
+        spec = CRUDSpec(self.model, params or {}, root_alias)
+        if params:
+            if self.supports_soft_delete:
+                include_deleted = _is_truthy(params.get('include_deleted'))
+        if self.supports_soft_delete and not include_deleted:
+            query = query.filter(getattr(root_alias, "is_deleted") == False)
+        query = query.filter(getattr(root_alias, "id") == id)
+
+        spec.parse_includes()
+
+        for parent_alias, relationship_attr, target_alias in spec.get_join_paths():
+            rel_attr = cast(InstrumentedAttribute, relationship_attr)
+            target = cast(Any, target_alias)
+            query = query.join(target, rel_attr.of_type(target), isouter=True)
+
+        if params:
+            root_fields, rel_field_names, root_field_names = spec.parse_fields()
+
+        fields = (params or {}).get("fields") if isinstance(params, dict) else None
+        if fields:
+            for opt in _loader_options_for_fields(root_alias, self.model, fields):
+                query = query.options(opt)
+
+        only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
+        if only_cols:
+            query = query.options(Load(root_alias).load_only(*only_cols))
+
+        for eager in spec.get_eager_loads(root_alias, fields_map=rel_field_names):
+            query = query.options(eager)
+
+        if params:
+            fields = params.get("fields") or []
+            for opt in _loader_options_for_fields(root_alias, self.model, fields):
+                query = query.options(opt)
+
+        obj = query.first()
+
+        proj = []
+        if root_field_names:
+            proj.extend(root_field_names)
+        if root_fields:
+            proj.extend(c.key for c in root_fields)
+        for path, names in (rel_field_names or {}).items():
+            prefix = ".".join(path)
+            for n in names:
+                proj.append(f"{prefix}.{n}")
+
+        if proj and "id" not in proj and hasattr(self.model, "id"):
+            proj.insert(0, "id")
+
+        if proj:
+            try:
+                setattr(obj, "__crudkit_projection__", tuple(proj))
+            except Exception:
+                pass
+
+        return obj or None
+
+    def list(self, params=None) -> list[T]:
+        query, root_alias = self.get_query()
+
+        root_fields = []
+        root_field_names = {}
+        rel_field_names = {}
+
+        if params:
+            if self.supports_soft_delete:
+                include_deleted = _is_truthy(params.get('include_deleted'))
+                if not include_deleted:
+                    query = query.filter(getattr(root_alias, "is_deleted") == False)
+
+            spec = CRUDSpec(self.model, params or {}, root_alias)
+            filters = spec.parse_filters()
+            order_by = spec.parse_sort()
+            limit, offset = spec.parse_pagination()
+            spec.parse_includes()
+
+            for parent_alias, relationship_attr, target_alias in spec.get_join_paths():
+                rel_attr = cast(InstrumentedAttribute, relationship_attr)
+                target = cast(Any, target_alias)
+                query = query.join(target, rel_attr.of_type(target), isouter=True)
+
+            if params:
+                root_fields, rel_field_names, root_field_names = spec.parse_fields()
+
+            only_cols = [c for c in root_fields if isinstance(c, InstrumentedAttribute)]
+            if only_cols:
+                query = query.options(Load(root_alias).load_only(*only_cols))
+
+            for eager in spec.get_eager_loads(root_alias, fields_map=rel_field_names):
+                query = query.options(eager)
+
+            if params:
+                fields = params.get("fields") or []
+                for opt in _loader_options_for_fields(root_alias, self.model, fields):
+                    query = query.options(opt)
+
+            if filters:
+                query = query.filter(*filters)
+
+            # MSSQL: requires ORDER BY when using OFFSET (and SQLA will use OFFSET for limit+offset).
+            paginating = (limit is not None) or (offset is not None and offset != 0)
+            if paginating and not order_by and self.backend.requires_order_by_for_offset:
+                order_by = self._default_order_by(root_alias)
+
+            if order_by:
+                query = query.order_by(*order_by)
+
+            # Only apply offset/limit when not None.
+            if offset is not None and offset != 0:
+                query = query.offset(offset)
+            if limit is not None and limit > 0:
+                query = query.limit(limit)
+
+        rows = query.all()
+
+        proj = []
+        if root_field_names:
+            proj.extend(root_field_names)
+        if root_fields:
+            proj.extend(c.key for c in root_fields)
+        for path, names in (rel_field_names or {}).items():
+            prefix = ".".join(path)
+            for n in names:
+                proj.append(f"{prefix}.{n}")
+
+        if proj and "id" not in proj and hasattr(self.model, "id"):
+            proj.insert(0, "id")
+
+        if proj:
+            for obj in rows:
+                try:
+                    setattr(obj, "__crudkit_projection__", tuple(proj))
+                except Exception:
+                    pass
+
+        return rows
+
+    def create(self, data: dict, actor=None) -> T:
+        obj = self.model(**data)
+        self.session.add(obj)
+        self.session.commit()
+        self._log_version("create", obj, actor)
+        return obj
+
+    def update(self, id: int, data: dict, actor=None) -> T:
+        obj = self.get(id)
+        if not obj:
+            raise ValueError(f"{self.model.__name__} with ID {id} not found.")
+        valid_fields = {c.name for c in self.model.__table__.columns}
+        for k, v in data.items():
+            if k in valid_fields:
+                setattr(obj, k, v)
+        self.session.commit()
+        self._log_version("update", obj, actor)
+        return obj
+
+    def delete(self, id: int, hard: bool = False, actor = False):
+        obj = self.session.get(self.model, id)
+        if not obj:
+            return None
+        if hard or not self.supports_soft_delete:
+            self.session.delete(obj)
+        else:
+            soft = cast(_SoftDeletable, obj)
+            soft.is_deleted = True
+        self.session.commit()
+        self._log_version("delete", obj, actor)
+        return obj
+
+    def _log_version(self, change_type: str, obj: T, actor=None, metadata: dict = {}):
+        try:
+            data = obj.as_dict()
+        except Exception:
+            data = {"error": "Failed to serialize object."}
+        version = Version(
+            model_name=self.model.__name__,
+            object_id=obj.id,
+            change_type=change_type,
+            data=data,
+            actor=str(actor) if actor else None,
+            meta=metadata
+        )
+        self.session.add(version)
+        self.session.commit()
diff --git a/crudkit/core/spec.py b/crudkit/core/spec.py
new file mode 100644
index 0000000..9c0e53b
--- /dev/null
+++ b/crudkit/core/spec.py
@@ -0,0 +1,189 @@
+from typing import List, Tuple, Set, Dict, Optional
+from sqlalchemy import asc, desc
+from sqlalchemy.ext.hybrid import hybrid_property
+from sqlalchemy.orm import aliased, selectinload
+from sqlalchemy.orm.attributes import InstrumentedAttribute
+
+OPERATORS = {
+    'eq': lambda col, val: col == val,
+    'lt': lambda col, val: col < val,
+    'lte': lambda col, val: col <= val,
+    'gt': lambda col, val: col > val,
+    'gte': lambda col, val: col >= val,
+    'ne': lambda col, val: col != val,
+    'icontains': lambda col, val: col.ilike(f"%{val}%"),
+}
+
+class CRUDSpec:
+    def __init__(self, model, params, root_alias):
+        self.model = model
+        self.params = params
+        self.root_alias = root_alias
+        self.eager_paths: Set[Tuple[str, ...]] = set()
+        self.join_paths: List[Tuple[object, InstrumentedAttribute, object]] = []
+        self.alias_map: Dict[Tuple[str, ...], object] = {}
+        self._root_fields: List[InstrumentedAttribute] = []
+        self._rel_field_names: Dict[Tuple[str, ...], object] = {}
+        self.include_paths: Set[Tuple[str, ...]] = set()
+
+    def _resolve_column(self, path: str):
+        current_alias = self.root_alias
+        parts = path.split('.')
+        join_path: list[str] = []
+
+        for i, attr in enumerate(parts):
+            try:
+                attr_obj = getattr(current_alias, attr)
+            except AttributeError:
+                return None, None
+
+            prop = getattr(attr_obj, "property", None)
+            if prop is not None and hasattr(prop, "direction"):
+                join_path.append(attr)
+                path_key = tuple(join_path)
+                alias = self.alias_map.get(path_key)
+                if not alias:
+                    alias = aliased(prop.mapper.class_)
+                    self.alias_map[path_key] = alias
+                    self.join_paths.append((current_alias, attr_obj, alias))
+                current_alias = alias
+                continue
+
+            if isinstance(attr_obj, InstrumentedAttribute) or getattr(attr_obj, "comparator", None) is not None or hasattr(attr_obj, "clauses"):
+                return attr_obj, tuple(join_path) if join_path else None
+
+        return None, None
+
+    def parse_includes(self):
+        raw = self.params.get('include')
+        if not raw:
+            return
+        tokens = [p.strip() for p in str(raw).split(',') if p.strip()]
+        for token in tokens:
+            _, join_path = self._resolve_column(token)
+            if join_path:
+                self.eager_paths.add(join_path)
+            else:
+                col, maybe = self._resolve_column(token + '.id')
+                if maybe:
+                    self.eager_paths.add(maybe)
+
+    def parse_filters(self):
+        filters = []
+        for key, value in self.params.items():
+            if key in ('sort', 'limit', 'offset'):
+                continue
+            if '__' in key:
+                path_op = key.rsplit('__', 1)
+                if len(path_op) != 2:
+                    continue
+                path, op = path_op
+            else:
+                path, op = key, 'eq'
+            col, join_path = self._resolve_column(path)
+            if col and op in OPERATORS:
+                filters.append(OPERATORS[op](col, value))
+                if join_path:
+                    self.eager_paths.add(join_path)
+        return filters
+
+    def parse_sort(self):
+        sort_args = self.params.get('sort', '')
+        result = []
+        for part in sort_args.split(','):
+            part = part.strip()
+            if not part:
+                continue
+            if part.startswith('-'):
+                field = part[1:]
+                order = desc
+            else:
+                field = part
+                order = asc
+            col, join_path = self._resolve_column(field)
+            if col:
+                result.append(order(col))
+                if join_path:
+                    self.eager_paths.add(join_path)
+        return result
+
+    def parse_pagination(self):
+        limit = int(self.params.get('limit', 100))
+        offset = int(self.params.get('offset', 0))
+        return limit, offset
+
+    def parse_fields(self):
+        """
+        Parse ?fields=colA,colB,rel1.colC,rel1.rel2.colD
+            - Root fields become InstrumentedAttributes bound to root_alias.
+            - Related fields store attribute NAMES; we'll resolve them on the target class when building loader options.
+        Returns (root_fields, rel_field_names).
+        """
+        raw = self.params.get('fields')
+        if not raw:
+            return [], {}, {}
+
+        if isinstance(raw, list):
+            tokens = []
+            for chunk in raw:
+                tokens.extend(p.strip() for p in str(chunk).split(',') if p.strip())
+        else:
+            tokens = [p.strip() for p in str(raw).split(',') if p.strip()]
+
+        root_fields: List[InstrumentedAttribute] = []
+        root_field_names: list[str] = []
+        rel_field_names: Dict[Tuple[str, ...], List[str]] = {}
+
+        for token in tokens:
+            col, join_path = self._resolve_column(token)
+            if not col:
+                continue
+            if join_path:
+                rel_field_names.setdefault(join_path, []).append(col.key)
+                self.eager_paths.add(join_path)
+            else:
+                root_fields.append(col)
+                root_field_names.append(getattr(col, "key", token))
+
+        seen = set()
+        root_fields = [c for c in root_fields if not (c.key in seen or seen.add(c.key))]
+        for k, names in rel_field_names.items():
+            seen2 = set()
+            rel_field_names[k] = [n for n in names if not (n in seen2 or seen2.add(n))]
+
+        self._root_fields = root_fields
+        self._rel_field_names = rel_field_names
+        return root_fields, rel_field_names, root_field_names
+
+    def get_eager_loads(self, root_alias, *, fields_map=None):
+        loads = []
+        for path in self.eager_paths:
+            current = root_alias
+            loader = None
+            for idx, name in enumerate(path):
+                rel_attr = getattr(current, name)
+                loader = selectinload(rel_attr) if loader is None else loader.selectinload(rel_attr)
+
+                # step to target class for the next hop
+                target_cls = rel_attr.property.mapper.class_
+                current = target_cls
+
+                # if final hop and we have a fields map, narrow columns
+                if fields_map and idx == len(path) - 1 and path in fields_map:
+                    cols = []
+                    for n in fields_map[path]:
+                        attr = getattr(target_cls, n, None)
+                        # Only include real column attributes; skip hybrids/expressions
+                        if isinstance(attr, InstrumentedAttribute):
+                            cols.append(attr)
+
+                    # Only apply load_only if we have at least one real column
+                    if cols:
+                        loader = loader.load_only(*cols)
+
+            if loader is not None:
+                loads.append(loader)
+        return loads
+
+    def get_join_paths(self):
+        return self.join_paths
diff --git a/crudkit/core/types.py b/crudkit/core/types.py
new file mode 100644
index 0000000..aade874
--- /dev/null
+++ b/crudkit/core/types.py
@@ -0,0 +1,16 @@
+from dataclasses import dataclass
+from typing import Any, Sequence
+
+@dataclass(frozen=True)
+class OrderSpec:
+    cols: Sequence[Any]
+    desc: Sequence[bool]
+
+@dataclass
+class SeekWindow:
+    items: list[Any]
+    limit: int
+    first_key: list[Any] | None
+    last_key: list[Any] | None
+    order: OrderSpec
+    total: int | None = None
diff --git a/crudkit/dsl.py b/crudkit/dsl.py
deleted file mode 100644
index 3953a83..0000000
--- a/crudkit/dsl.py
+++ /dev/null
@@ -1,147 +0,0 @@
-from dataclasses import dataclass, field
-from typing import List, Dict, Any, Optional
-from sqlalchemy import asc, desc, select, false
-from sqlalchemy.inspection import inspect
-
-@dataclass
-class QuerySpec:
-    filters: Dict[str, Any] = field(default_factory=dict)
-    order_by: List[str] = field(default_factory=list)
-    page: Optional[int] = None
-    per_page: Optional[int] = None
-    expand: List[str] = field(default_factory=list)
-    fields: Optional[List[str]] = None
-
-FILTER_OPS = {
-    "__eq": lambda c, v: c == v,
-    "__ne": lambda c, v: c != v,
-    "__lt": lambda c, v: c < v,
-    "__lte": lambda c, v: c <= v,
-    "__gt": lambda c, v: c > v,
-    "__gte": lambda c, v: c >= v,
-    "__ilike": lambda c, v: c.ilike(v),
-    "__in": lambda c, v: c.in_(v),
-    "__isnull": lambda c, v: (c.is_(None) if v else c.is_not(None))
-}
-
-def _split_filter_key(raw_key: str):
-    for op in sorted(FILTER_OPS.keys(), key=len, reverse=True):
-        if raw_key.endswith(op):
-            return raw_key[: -len(op)], op
-    return raw_key, None
-
-def _ensure_wildcards(op_key, value):
-    if op_key == "__ilike" and isinstance(value, str) and "%" not in value and "_" not in value:
-        return f"%{value}%"
-    return value
-
-def _related_predicate(Model, path_parts, op_key, value):
-    """
-    Build EXISTS subqueries for dotted filters:
-      - scalar rels  -> attr.has(inner_predicate)
-      - collection   -> attr.any(inner_predicate)
-    """
-    head, *rest = path_parts
-
-    # class-bound relationship attribute (InstrumentedAttribute)
-    attr = getattr(Model, head, None)
-    if attr is None:
-        return None
-
-    # relationship metadata if you need uselist + target model
-    rel = inspect(Model).relationships.get(head)
-    if rel is None:
-        return None
-    Target = rel.mapper.class_
-
-    if not rest:
-        # filtering directly on a relationship without a leaf column isn't supported
-        return None
-
-    if len(rest) == 1:
-        # final hop is a column on the related model
-        leaf = rest[0]
-        col = getattr(Target, leaf, None)
-        if col is None:
-            return None
-        pred = FILTER_OPS[op_key](col, value) if op_key else (col == value)
-    else:
-        # recurse deeper: owner.room.area.name__ilike=...
-        pred = _related_predicate(Target, rest, op_key, value)
-        if pred is None:
-            return None
-
-    # wrap at this hop using the *attribute*, not the RelationshipProperty
-    return attr.any(pred) if rel.uselist else attr.has(pred)
-
-def split_sort_tokens(tokens):
-    simple, dotted = [], []
-    for tok in (tokens or []):
-        if not tok:
-            continue
-        key = tok.lstrip("-")
-        if ":" in key:
-            key = key.split(":", 1)[0]
-        (dotted if "." in key else simple).append(tok)
-    return simple, dotted
-
-def build_query(Model, spec: QuerySpec, eager_policy=None):
-    stmt = select(Model)
-
-    # filter out soft-deleted rows
-    deleted_attr = getattr(Model, "deleted", None)
-    if deleted_attr is not None:
-        stmt = stmt.where(deleted_attr == false())
-    else:
-        is_deleted_attr = getattr(Model, "is_deleted", None)
-        if is_deleted_attr is not None:
-            stmt = stmt.where(is_deleted_attr == false())
-
-    # filters
-    for raw_key, val in spec.filters.items():
-        path, op_key = _split_filter_key(raw_key)
-        val = _ensure_wildcards(op_key, val)
-
-        if "." in path:
-            pred = _related_predicate(Model, path.split("."), op_key, val)
-            if pred is not None:
-                stmt = stmt.where(pred)
-            continue
-
-        col = getattr(Model, path, None)
-        if col is None:
-            continue
-        stmt = stmt.where(FILTER_OPS[op_key](col, val) if op_key else (col == val))
-
-    simple_sorts, _ = split_sort_tokens(spec.order_by)
-
-    for token in simple_sorts:
-        direction = "asc"
-        key = token
-        if token.startswith("-"):
-            direction = "desc"
-            key = token[1:]
-        if ":" in key:
-            key, d = key.rsplit(":", 1)
-            direction = "desc" if d.lower().startswith("d") else "asc"
-
-        if "." in key:
-            continue
-
-        col = getattr(Model, key, None)
-        if col is None:
-            continue
-        stmt = stmt.order_by(desc(col) if direction == "desc" else asc(col))
-
-    if not spec.order_by and spec.page and spec.per_page:
-        pk_cols = inspect(Model).primary_key
-        if pk_cols:
-            stmt = stmt.order_by(*(asc(c) for c in pk_cols))
-
-    # eager loading
-    if eager_policy:
-        opts = eager_policy(Model, spec.expand)
-        if opts:
-            stmt = stmt.options(*opts)
-
-    return stmt
diff --git a/crudkit/eager.py b/crudkit/eager.py
deleted file mode 100644
index 34e7884..0000000
--- a/crudkit/eager.py
+++ /dev/null
@@ -1,75 +0,0 @@
-from __future__ import annotations
-from typing import Iterable, List, Sequence, Set
-from sqlalchemy.inspection import inspect
-from sqlalchemy.orm import Load, joinedload, selectinload, RelationshipProperty
-
-class EagerConfig:
-    def __init__(self, strict: bool = False, max_depth: int = 4):
-        self.strict = strict
-        self.max_depth = max_depth
-
-def _rel(cls, name: str) -> RelationshipProperty | None:
-    return inspect(cls).relationships.get(name)
-
-def _is_expandable(rel: RelationshipProperty) -> bool:
-    # Skip dynamic or viewonly collections; they don’t support eagerload
-    return rel.lazy != "dynamic"
-
-def default_eager_policy(Model, expand: Sequence[str], cfg: EagerConfig | None = None) -> List[Load]:
-    """
-    Heuristic:
-      - many-to-one / one-to-one: joinedload
-      - one-to-many / many-to-many: selectinload
-    Accepts dotted paths like "author.publisher".
-    """
-    if not expand:
-        return []
-
-    cfg = cfg or EagerConfig()
-    # normalize, dedupe, and prefer longer paths over their prefixes
-    raw: Set[str] = {p.strip() for p in expand if p and p.strip()}
-    # drop prefixes if a longer path exists (author, author.publisher -> keep only author.publisher)
-    pruned: Set[str] = set(raw)
-    for p in raw:
-        parts = p.split(".")
-        for i in range(1, len(parts)):
-            pruned.discard(".".join(parts[:i]))
-
-    opts: List[Load] = []
-    seen: Set[tuple] = set()
-
-    for path in sorted(pruned):
-        parts = path.split(".")
-        if len(parts) > cfg.max_depth:
-            if cfg.strict:
-                raise ValueError(f"expand path too deep: {path} (max {cfg.max_depth})")
-            continue
-
-        current_model = Model
-        # build the chain incrementally
-        loader: Load | None = None
-        ok = True
-
-        for i, name in enumerate(parts):
-            rel = _rel(current_model, name)
-            if not rel or not _is_expandable(rel):
-                ok = False
-                break
-            attr = getattr(current_model, name)
-            if loader is None:
-                loader = selectinload(attr) if rel.uselist else joinedload(attr)
-            else:
-                loader = loader.selectinload(attr) if rel.uselist else loader.joinedload(attr)
-            current_model = rel.mapper.class_
-
-        if not ok:
-            if cfg.strict:
-                raise ValueError(f"unknown or non-expandable relationship in expand path: {path}")
-            continue
-
-        key = (tuple(parts),)
-        if loader is not None and key not in seen:
-            opts.append(loader)
-            seen.add(key)
-
-    return opts
diff --git a/crudkit/engines.py b/crudkit/engines.py
new file mode 100644
index 0000000..b420a8d
--- /dev/null
+++ b/crudkit/engines.py
@@ -0,0 +1,50 @@
+from __future__ import annotations
+from typing import Type, Optional
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+from .backend import make_backend_info, BackendInfo
+from .config import Config, get_config
+from ._sqlite import apply_sqlite_pragmas
+
+def build_engine(config_cls: Type[Config] | None = None):
+    config_cls = config_cls or get_config(None)
+    engine = create_engine(config_cls.DATABASE_URL, **config_cls.engine_kwargs())
+    apply_sqlite_pragmas(engine, config_cls.SQLITE_PRAGMAS)
+    return engine
+
+def build_sessionmaker(config_cls: Type[Config] | None = None, engine=None):
+    config_cls = config_cls or get_config(None)
+    engine = engine or build_engine(config_cls)
+    return sessionmaker(bind=engine, **config_cls.session_kwargs())
+
+class CRUDKitRuntime:
+    """
+    Lightweight container so CRUDKit can be given either:
+        - prebuild engine/sessionmaker, or
+        - a Config to build them lazily
+    """
+    def __init__(self, *, engine=None, session_factory=None, config: Optional[Type[Config]] = None):
+        if engine is None and session_factory is None and config is None:
+            config = get_config(None)
+        self._config = config
+        self._engine = engine or (build_engine(config) if config else None)
+        self._session_factory = session_factory or (build_sessionmaker(config, self._engine) if config else None)
+
+    @property
+    def engine(self):
+        if self._engine is None and self._config:
+            self._engine = build_engine(self._config)
+        return self._engine
+
+    @property
+    def session_factory(self):
+        if self._session_factory is None:
+            if self._config and self._engine:
+                self._session_factory = build_sessionmaker(self._config, self._engine)
+        return self._session_factory
+
+    @property
+    def backend(self) -> BackendInfo:
+        if not hasattr(self, "_backend_info") or self._backend_info is None:
+            self._backend_info = make_backend_info(self.engine)
+        return self._backend_info
diff --git a/crudkit/html/__init__.py b/crudkit/html/__init__.py
deleted file mode 100644
index a94f018..0000000
--- a/crudkit/html/__init__.py
+++ /dev/null
@@ -1,3 +0,0 @@
-from .ui_fragments import make_fragments_blueprint
-
-__all__ = ["make_fragments_blueprint"]
diff --git a/crudkit/html/templates/crudkit/_macros.html b/crudkit/html/templates/crudkit/_macros.html
deleted file mode 100644
index 342c731..0000000
--- a/crudkit/html/templates/crudkit/_macros.html
+++ /dev/null
@@ -1,140 +0,0 @@
-{% macro options(items, value_attr="id", label_path="name", getp=None) -%}
-
-{%- for obj in items -%}
-<option value="{{ getp(obj, value_attr) }}">{{ getp(obj, label_path) }}</option>
-{%- endfor -%}
-{% endmacro %}
-
-{% macro lis(items, label_path="name", sublabel_path=None, getp=None) -%}
-{%- for obj in items -%}
-<li data-id="{{ obj.id }}">
-    <div class="li-main">{{ getp(obj, label_path) }}</div>
-    {%- if sublabel_path %}
-    <div class="li-sub">{{ getp(obj, sublabel_path) }}</div>
-    {%- endif %}
-</li>
-{%- else -%}
-<li class="empty"><em>No results.</em></li>
-{%- endfor -%}
-{% endmacro %}
-
-{% macro rows(items, fields, getp=None) -%}
-{%- for obj in items -%}
-<tr id="row-{{ obj.id }}">
-    {%- for f in fields -%}
-    <td data-field="{{ f }}">{{ getp(obj, f) }}</td>
-    {%- endfor -%}
-</tr>
-{%- else -%}
-<tr>
-    <td colspan="{{ fields|length }}"><em>No results.</em></td>
-</tr>
-{%- endfor -%}
-{%- endmacro %}
-
-{# helper: centralize the query string once #}
-{% macro _q(model, page, per_page, sort, filters, fields_csv) -%}
-/ui/{{ model }}/frag/rows
-?page={{ page }}&per_page={{ per_page }}
-{%- if sort %}&sort={{ sort }}{% endif -%}
-{%- if fields_csv %}&fields_csv={{ fields_csv|urlencode }}{% endif -%}
-{%- for k, v in (filters or {}).items() %}&{{ k }}={{ v|urlencode }}{% endfor -%}
-{%- endmacro %}
-
-{% macro pager(model, page, pages, per_page, sort, filters, fields_csv) -%}
-{% set p = page|int %}
-{% set pg = pages|int %}
-{% set prev = 1 if p <= 1 else p - 1 %} {% set nxt=pg if p>= pg else p + 1 %}
-
-    <nav class="pager-nav" aria-label="Pagination">
-        {% if p > 1 %}
-        <button type="button" class="page-btn" data-page="1"
-            hx-get="{{ _q(model, 1, per_page, sort, filters, fields_csv) }}" hx-target="#rows" hx-swap="innerHTML"
-            aria-label="First page">First</button>
-
-        <button type="button" class="page-btn" data-page="{{ prev }}"
-            hx-get="{{ _q(model, prev, per_page, sort, filters, fields_csv) }}" hx-target="#rows" hx-swap="innerHTML"
-            rel="prev">Prev</button>
-        {% else %}
-        <button type="button" class="page-btn" disabled>First</button>
-        <button type="button" class="page-btn" disabled>Prev</button>
-        {% endif %}
-
-        <span aria-live="polite">Page {{ p }} / {{ pg }}</span>
-
-        {% if p < pg %} <button type="button" class="page-btn" data-page="{{ p + 1 }}"
-            hx-get="{{ _q(model, p + 1, per_page, sort, filters, fields_csv) }}" hx-target="#rows" hx-swap="innerHTML"
-            rel="next">Next</button>
-
-            <button type="button" class="page-btn" data-page="{{ pg }}"
-                hx-get="{{ _q(model, pg, per_page, sort, filters, fields_csv) }}" hx-target="#rows" hx-swap="innerHTML"
-                aria-label="Last page">Last</button>
-            {% else %}
-            <button type="button" class="page-btn" disabled>Next</button>
-            <button type="button" class="page-btn" disabled>Last</button>
-            {% endif %}
-    </nav>
-
-    {# one tiny listener to keep #pager-state in sync for every button #}
-    <script>
-        (function () {
-            const nav = document.currentScript.previousElementSibling;
-            nav.addEventListener('click', function (ev) {
-                const btn = ev.target.closest('.page-btn');
-                if (!btn || btn.disabled) return;
-                const page = btn.getAttribute('data-page');
-                if (!page) return;
-                const inp = document.querySelector('#pager-state input[name=page]');
-                if (inp) inp.value = page;
-            }, { capture: true });
-        })();
-    </script>
-    {%- endmacro %}
-
-    {% macro form(schema, action, method="POST", obj_id=None, hx=False, csrf_token=None) -%}
-    <form action="{{ action }}" method="post" {%- if hx %} hx-{{ "patch" if obj_id else "post" }}="{{ action }}"
-        hx-target="closest dialog, #modal-body, body" hx-swap="innerHTML" hx-disabled-elt="button[type=submit]" {%-
-        endif -%}>
-        {%- if csrf_token %}<input type="hidden" name="csrf_token" value="{{ csrf_token() }}">{% endif -%}
-        {%- if obj_id %}<input type="hidden" name="id" value="{{ obj_id }}">{% endif -%}
-        <input type="hidden" name="fields_csv" value="{{ request.args.get('fields_csv','id,name') }}">
-
-        {%- for f in schema -%}
-        <div class="field" data-name="{{ f.name }}">
-            {% set fid = 'f-' ~ f.name ~ '-' ~ (obj_id or 'new') %}
-            <label for="{{ fid }}">{{ f.label or f.name|replace('_',' ')|title }}</label>
-            {%- if f.type == "textarea" -%}
-            <textarea id="{{ fid }}" name="{{ f.name }}" {%- if f.required %} required{% endif %}{% if f.maxlength %}
-                maxlength="{{ f.maxlength }}" {% endif %}>{{ f.value or "" }}</textarea>
-            {%- elif f.type == "select" -%}
-            <select id="{{ fid }}" name="{{ f.name }}" {% if f.required %}required{% endif %}>
-                <option value="">{{ f.placeholder or ("Choose " ~ (f.label or f.name|replace('_',' ')|title)) }}
-                </option>
-                {% if f.multiple %}
-                {% set selected = (f.value or [])|list %}
-                {% for val, lbl in f.choices %}
-                <option value="{{ val }}" {{ 'selected' if val in selected else '' }}>{{ lbl }}</option>
-                {% endfor %}
-                {% else %}
-                {% for val, lbl in f.choices %}
-                <option value="{{ val }}" {{ 'selected' if (f.value|string)==(val|string) else '' }}>{{ lbl }}</option>
-                {% endfor %}
-                {% endif %}
-            </select>
-            {%- elif f.type == "checkbox" -%}
-            <input type="hidden" name="{{ f.name }}" value="0">
-            <input id="{{ fid }}" type="checkbox" name="{{ f.name }}" value="1" {{ "checked" if f.value else "" }}>
-            {%- else -%}
-            <input id="{{ fid }}" type="{{ f.type }}" name="{{ f.name }}"
-                value="{{ f.value if f.value is not none else '' }}" {%- if f.required %} required{% endif %} {%- if
-                f.maxlength %} maxlength="{{ f.maxlength }}" {% endif %}>
-            {%- endif -%}
-            {%- if f.help %}<div class="help">{{ f.help }}</div>{% endif -%}
-        </div>
-        {%- endfor -%}
-
-        <div class="actions">
-            <button type="submit">Save</button>
-        </div>
-    </form>
-    {%- endmacro %}
\ No newline at end of file
diff --git a/crudkit/html/templates/crudkit/form.html b/crudkit/html/templates/crudkit/form.html
deleted file mode 100644
index 761658a..0000000
--- a/crudkit/html/templates/crudkit/form.html
+++ /dev/null
@@ -1,3 +0,0 @@
-{% import "crudkit/_macros.html" as ui %}
-{% set action = url_for('frags.save', model=model) %}
-{{ ui.form(schema, action, method="POST", obj_id=obj.id if obj else None, hx=true) }}
\ No newline at end of file
diff --git a/crudkit/html/templates/crudkit/lis.html b/crudkit/html/templates/crudkit/lis.html
deleted file mode 100644
index 3ce62e7..0000000
--- a/crudkit/html/templates/crudkit/lis.html
+++ /dev/null
@@ -1,2 +0,0 @@
-{% import "crudkit/_macros.html" as ui %}
-{{ ui.lis(items, label_path=label_path, sublabel_path=sublabel_path, getp=getp) }}
diff --git a/crudkit/html/templates/crudkit/options.html b/crudkit/html/templates/crudkit/options.html
deleted file mode 100644
index 0f66d2d..0000000
--- a/crudkit/html/templates/crudkit/options.html
+++ /dev/null
@@ -1,3 +0,0 @@
-{# Renders only <option>...</option> rows #}
-{% import "crudkit/_macros.html" as ui %}
-{{ ui.options(items, value_attr=value_attr, label_path=label_path, getp=getp) }}
diff --git a/crudkit/html/templates/crudkit/pager.html b/crudkit/html/templates/crudkit/pager.html
deleted file mode 100644
index b3e22a3..0000000
--- a/crudkit/html/templates/crudkit/pager.html
+++ /dev/null
@@ -1,2 +0,0 @@
-{% import 'crudkit/_macros.html' as ui %}
-{{ ui.pager(model, page, pages, per_page, sort, filters, fields_csv) }}
diff --git a/crudkit/html/templates/crudkit/row.html b/crudkit/html/templates/crudkit/row.html
deleted file mode 100644
index 719c0ba..0000000
--- a/crudkit/html/templates/crudkit/row.html
+++ /dev/null
@@ -1,2 +0,0 @@
-{% import "crudkit/_macros.html" as ui %}
-{{ ui.rows([obj], fields, getp=getp) }}
\ No newline at end of file
diff --git a/crudkit/html/templates/crudkit/rows.html b/crudkit/html/templates/crudkit/rows.html
deleted file mode 100644
index 7784328..0000000
--- a/crudkit/html/templates/crudkit/rows.html
+++ /dev/null
@@ -1,2 +0,0 @@
-{% import "crudkit/_macros.html" as ui %}
-{{ ui.rows(items, fields, getp=getp) }}
diff --git a/crudkit/html/type_map.py b/crudkit/html/type_map.py
deleted file mode 100644
index 5e582c2..0000000
--- a/crudkit/html/type_map.py
+++ /dev/null
@@ -1,137 +0,0 @@
-from __future__ import annotations
-from typing import Any, Dict, List, Optional, Tuple
-from sqlalchemy import select
-from sqlalchemy.inspection import inspect
-from sqlalchemy.orm import Mapper, RelationshipProperty
-from sqlalchemy.sql.schema import Column
-from sqlalchemy.sql.sqltypes import (
-    String, Text, Unicode, UnicodeText,
-    Integer, BigInteger, SmallInteger, Float, Numeric, Boolean,
-    Date, DateTime, Time, JSON, Enum
-)
-
-CANDIDATE_LABELS = ("name", "title", "label", "display_name")
-
-def _guess_label_attr(model_cls) -> str:
-    for cand in CANDIDATE_LABELS:
-        if hasattr(model_cls, cand):
-            return cand
-    return "id"
-
-def _pretty(label: str) -> str:
-    return label.replace("_", " ").title()
-
-def _column_input_type(col: Column) -> str:
-    t = col.type
-    if isinstance(t, (String, Unicode)):
-        return "text"
-    if isinstance(t, (Text, UnicodeText, JSON)):
-        return "textarea"
-    if isinstance(t, (Integer, SmallInteger, BigInteger)):
-        return "number"
-    if isinstance(t, (Float, Numeric)):
-        return "number"
-    if isinstance(t, Boolean):
-        return "checkbox"
-    if isinstance(t, Date):
-        return "date"
-    if isinstance(t, DateTime):
-        return "datetime-local"
-    if isinstance(t, Time):
-        return "time"
-    if isinstance(t, Enum):
-        return "select"
-    return "text"
-
-def _enum_choices(col: Column) -> Optional[List[Tuple[str, str]]]:
-    t = col.type
-    if isinstance(t, Enum):
-        if t.enum_class:
-            return [(e.name, e.value) for e in t.enum_class]
-        if t.enums:
-            return [(v, v) for v in t.enums]
-    return None
-
-def build_form_schema(model_cls, session, obj=None, *, include=None, exclude=None, fk_limit=200):
-    mapper: Mapper = inspect(model_cls)
-    include = set(include or [])
-    exclude = set(exclude or {"id", "created_at", "updated_at", "deleted", "version"})
-    fields = []
-
-    fields: List[Dict[str, Any]] = []
-
-    fk_map = {}
-    for rel in mapper.relationships:
-        for lc in rel.local_columns:
-            fk_map[lc.key] = rel
-
-    for attr in mapper.column_attrs:
-        col = attr.columns[0]
-        name = col.key
-        if include and name not in include:
-            continue
-        if name in exclude:
-            continue
-
-        field = {
-            "name": name,
-            "type": _column_input_type(col),
-            "required": not col.nullable,
-            "value": getattr(obj, name, None) if obj is not None else None,
-            "placeholder": "",
-            "help": "",
-            # default label from column name
-            "label": _pretty(name),
-        }
-
-        enum_choices = _enum_choices(col)
-        if enum_choices:
-            field["type"] = "select"
-            field["choices"] = enum_choices
-
-        if name in fk_map:
-            rel = fk_map[name]
-            target = rel.mapper.class_
-            label_attr = _guess_label_attr(target)
-            rows = session.execute(select(target).limit(fk_limit)).scalars().all()
-            field["type"] = "select"
-            field["choices"] = [(getattr(r, "id"), getattr(r, label_attr)) for r in rows]
-            field["rel"] = {"target": target.__name__, "label_attr": label_attr}
-            field["label"] = _pretty(rel.key)
-
-        if getattr(col.type, "length", None):
-            field["maxlength"] = col.type.length
-
-        fields.append(field)
-
-    for rel in mapper.relationships:
-        if not rel.uselist or rel.secondary is None:
-            continue  # only true many-to-many
-
-        if include and f"{rel.key}_ids" not in include:
-            continue
-
-        target = rel.mapper.class_
-        label_attr = _guess_label_attr(target)
-        choices = session.execute(select(target).limit(fk_limit)).scalars().all()
-
-        current = []
-        if obj is not None:
-            current = [getattr(x, "id") for x in getattr(obj, rel.key, []) or []]
-
-        fields.append({
-            "name": f"{rel.key}_ids",     # e.g. "tags_ids"
-            "label": rel.key.replace("_"," ").title(),
-            "type": "select",
-            "multiple": True,
-            "required": False,
-            "choices": [(getattr(r,"id"), getattr(r,label_attr)) for r in choices],
-            "value": current,             # list of selected IDs
-            "placeholder": f"Choose {rel.key.replace('_',' ').title()}",
-            "help": "",
-        })
-
-    if include:
-        order = list(include)
-        fields.sort(key=lambda f: order.index(f["name"]) if f["name"] in include else 10**9)
-    return fields
diff --git a/crudkit/html/ui_fragments.py b/crudkit/html/ui_fragments.py
deleted file mode 100644
index c2da1da..0000000
--- a/crudkit/html/ui_fragments.py
+++ /dev/null
@@ -1,269 +0,0 @@
-from __future__ import annotations
-from typing import Any, Dict, List, Tuple
-from math import ceil
-from flask import Blueprint, request, render_template, abort, make_response
-from sqlalchemy import select
-from sqlalchemy.orm import scoped_session
-from sqlalchemy.inspection import inspect
-from sqlalchemy.sql.elements import UnaryExpression
-from sqlalchemy.sql.sqltypes import Integer, Boolean, Date, DateTime, Float, Numeric
-
-from ..dsl import QuerySpec
-from ..service import CrudService
-from ..eager import default_eager_policy
-from .type_map import build_form_schema
-
-Session = None
-
-def make_fragments_blueprint(db_session_factory, registry: Dict[str, Any], *, name="frags"):
-    """
-    HTML fragments for HTMX/Alpine. No base pages. Pure partials:
-        GET /<model>/frag/options   -> <option>...</option>
-        GET /<model>/frag/lis       -> <li>...</li>
-        GET /<model>/frag/rows      -> <tr>...</tr> + pager markup if wanted
-        GET /<model>/frag/form      -> <form>...</form> (auto-generated)
-    """
-    global Session
-    if Session is None:
-        Session = scoped_session(db_session_factory)
-
-    bp = Blueprint(name, __name__, template_folder="templates")
-
-    def session():
-        return Session
-
-    @bp.teardown_app_request
-    def remove_session(exc=None):
-        Session.remove()
-
-    def _parse_filters(args):
-        reserved = {"page", "per_page", "sort", "expand", "fields", "value", "label", "label_tpl", "fields_csv", "li_label", "li_sublabel"}
-        out = {}
-        for k, v in args.items():
-            if k not in reserved and v != "":
-                out[k] = v
-        return out
-
-    def _paths_from_csv(csv: str) -> List[str]:
-        return [p.strip() for p in csv.split(",") if p.strip()]
-
-    def _collect_expand_from_paths(paths: List[str]) -> List[str]:
-        rels = set()
-        for p in paths:
-            bits = p.split(".")
-            if len(bits) > 1:
-                rels.add(bits[0])
-        return list(rels)
-
-    def _getp(obj, path: str):
-        cur = obj
-        for part in path.split("."):
-            cur = getattr(cur, part, None) if cur is not None else None
-        return cur
-
-    def _extract_m2m_lists(Model, req_form) -> dict[str, list[int]]:
-        """Return {'tags': [1,2]} for any <rel>_ids fields; caller removes keys from main form."""
-        mapper = inspect(Model)
-        out = {}
-        for rel in mapper.relationships:
-            if not rel.uselist or rel.secondary is None:
-                continue
-            key = f"{rel.key}_ids"
-            ids = req_form.getlist(key)
-            if ids is None:
-                continue
-            out[rel.key] = [int(i) for i in ids if i]
-        return out
-
-    @bp.get("/<model>/frag/options")
-    def options(model):
-        Model = registry.get(model) or abort(404)
-        value_attr = request.args.get("value", default="id")
-        label_path = request.args.get("label", default="name")
-        filters = _parse_filters(request.args)
-
-        expand = _collect_expand_from_paths([label_path])
-        spec = QuerySpec(filters=filters, order_by=[], page=None, per_page=None, expand=expand)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        items, _ = svc.list(Model, spec)
-
-        return render_template("crudkit/options.html", items=items, value_attr=value_attr, label_path=label_path, getp=_getp)
-
-    @bp.get("/<model>/frag/lis")
-    def lis(model):
-        Model = registry.get(model) or abort(404)
-        label_path = request.args.get("li_label", default="name")
-        sublabel_path = request.args.get("li_sublabel")
-        filters = _parse_filters(request.args)
-        sort = request.args.get("sort")
-        page = request.args.get("page", type=int)
-        per_page = request.args.get("per_page", type=int)
-
-        expand = _collect_expand_from_paths([p for p in (label_path, sublabel_path) if p])
-        spec = QuerySpec(filters=filters, order_by=[sort] if sort else [], page=page, per_page=per_page, expand=expand)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        rows, total = svc.list(Model, spec)
-        pages = (ceil(total / per_page) if page and per_page else 1)
-        return render_template("crudkit/lis.html", items=rows, label_path=label_path, sublabel_path=sublabel_path, page=page or 1, per_page=per_page or 1, total=total, model=model, sort=sort, filters=filters, getp=_getp)
-
-    @bp.get("/<model>/frag/rows")
-    def rows(model):
-        Model = registry.get(model) or abort(404)
-        fields_csv = request.args.get("fields_csv") or "id,name"
-        fields = _paths_from_csv(fields_csv)
-        filters = _parse_filters(request.args)
-        sort = request.args.get("sort")
-        page = request.args.get("page", type=int) or 1
-        per_page = request.args.get("per_page", type=int) or 20
-
-        expand = _collect_expand_from_paths(fields + ([sort.split(":")[0]] if sort else []))
-        spec = QuerySpec(filters=filters, order_by=[sort] if sort else [], page=page, per_page=per_page, expand=expand)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        rows, _ = svc.list(Model, spec)
-
-        html = render_template("crudkit/rows.html", items=rows, fields=fields, getp=_getp, model=model)
-
-        return html
-
-    @bp.get("/<model>/frag/pager")
-    def pager(model):
-        Model = registry.get(model) or abort(404)
-        page = request.args.get("page", type=int) or 1
-        print(page)
-        per_page = request.args.get("per_page", type=int) or 20
-        filters = _parse_filters(request.args)
-        sort = request.args.get("sort")
-        fields_csv = request.args.get("fields_csv") or "id,name"
-        fields = _paths_from_csv(fields_csv)
-        expand = _collect_expand_from_paths(fields + ([sort.split(":")[0]] if sort else []))
-
-        spec = QuerySpec(filters=filters, order_by=[sort] if sort else [], page=page, per_page=per_page, expand=expand)
-        s = session(); svc = CrudService(s, default_eager_policy)
-        _, total = svc.list(Model, spec)
-        pages = max(1, ceil(total / per_page))
-
-        html = render_template("crudkit/pager.html", model=model, page=page, pages=pages,
-                                     per_page=per_page, sort=sort, filters=filters, fields_csv=fields_csv)
-        return html
-
-    @bp.get("/<model>/frag/form")
-    def form(model):
-        Model = registry.get(model) or abort(404)
-        id = request.args.get("id", type=int)
-        include_csv = request.args.get("include")
-        include = [s.strip() for s in include_csv.split(",")] if include_csv else None
-
-        s = session(); svc = CrudService(s, default_eager_policy)
-        obj = svc.get(Model, id) if id else None
-
-        schema = build_form_schema(Model, s, obj=obj, include=include)
-
-        hx = request.args.get("hx", type=int) == 1
-        return render_template("crudkit/form.html", model=model, obj=obj, schema=schema, hx=hx)
-
-    def coerce_form_types(Model, data: dict) -> dict:
-        """Turn HTML string inputs into the Python types your columns expect."""
-        mapper = inspect(Model)
-        for attr in mapper.column_attrs:
-            col = attr.columns[0]
-            name = col.key
-            if name not in data:
-                continue
-            v = data[name]
-            if v == "":
-                data[name] = None
-                continue
-            t = col.type
-            try:
-                if isinstance(t, Boolean):
-                    data[name] = v in ("1", "true", "on", "yes", True)
-                elif isinstance(t, Integer):
-                    data[name] = int(v)
-                elif isinstance(t, (Float, Numeric)):
-                    data[name] = float(v)
-                elif isinstance(t, DateTime):
-                    from datetime import datetime
-                    data[name] = datetime.fromisoformat(v)
-                elif isinstance(t, Date):
-                    from datetime import date
-                    data[name] = date.fromisoformat(v)
-            except Exception:
-                # Leave as string; your validator can complain later.
-                pass
-        return data
-
-    @bp.post("/<model>/frag/save")
-    def save(model):
-        Model = registry.get(model) or abort(404)
-        s = session(); svc = CrudService(s, default_eager_policy)
-
-        # grab the raw form and fields to re-render
-        raw = request.form
-        form = raw.to_dict(flat=True)
-        fields_csv = form.pop("fields_csv", "id,name")
-
-        # many-to-many lists first
-        m2m = _extract_m2m_lists(Model, raw)
-        for rel_name in list(m2m.keys()):
-            form.pop(f"{rel_name}_ids", None)
-
-        # coerce primitives for regular columns
-        form = coerce_form_types(Model, form)
-
-        id_val = form.pop("id", None)
-
-        if id_val:
-            obj = svc.get(Model, int(id_val)) or abort(404)
-            svc.update(obj, form)
-        else:
-            obj = svc.create(Model, form)
-
-        # apply many-to-many selections
-        mapper = inspect(Model)
-        for rel_name, id_list in m2m.items():
-            rel = mapper.relationships[rel_name]
-            target = rel.mapper.class_
-            selected = []
-            if id_list:
-                selected = s.execute(select(target).where(target.id.in_(id_list))).scalars().all()
-            coll = getattr(obj, rel_name)
-            coll.clear()
-            coll.extend(selected)
-
-        s.commit()
-
-        rows_html = render_template(
-            "crudkit/row.html",
-            obj=obj,
-            fields=[p.strip() for p in fields_csv.split(",") if p.strip()],
-            getp=_getp,
-        )
-        resp = make_response(rows_html)
-        if id_val:
-            resp.headers["HX-Trigger"] = '{"toast":{"level":"success","message":"Updated"}}'
-            resp.headers["HX-Retarget"] = f"#row-{obj.id}"
-            resp.headers["HX-Reswap"] = "outerHTML"
-        else:
-            resp.headers["HX-Trigger"] = '{"toast":{"level":"success","message":"Created"}}'
-            resp.headers["HX-Retarget"] = "#rows"
-            resp.headers["HX-Reswap"] = "beforeend"
-        return resp
-
-    @bp.get("/_debug/<model>/schema")
-    def debug_model(model):
-        Model = registry[model]
-        from sqlalchemy.inspection import inspect
-        m = inspect(Model)
-        return {
-            "columns": [c.key for c in m.columns],
-            "relationships": [
-                {
-                    "key": r.key,
-                    "target": r.mapper.class_.__name__,
-                    "uselist": r.uselist,
-                    "local_cols": [c.key for c in r.local_columns],
-                } for r in m.relationships
-            ],
-        }
-    return bp
-
diff --git a/crudkit/integration.py b/crudkit/integration.py
new file mode 100644
index 0000000..665e813
--- /dev/null
+++ b/crudkit/integration.py
@@ -0,0 +1,25 @@
+from __future__ import annotations
+from typing import Type
+from flask import Flask
+
+from crudkit.engines import CRUDKitRuntime
+
+from .registry import CRUDRegistry
+
+class CRUDKit:
+    def __init__(self, app: Flask, runtime: CRUDKitRuntime):
+        self.app = app
+        self.runtime = runtime
+        self.registry = CRUDRegistry(runtime)
+
+    def register(self, model: Type, **kwargs):
+        return self.registry.register_class(self.app, model, **kwargs)
+
+    def register_many(self, models: list[Type], **kwargs):
+        return self.registry.register_many(self.app, models, **kwargs)
+
+    def get_model(self, key: str):
+        return self.registry.get_model(key)
+
+    def get_service(self, model: Type):
+        return self.registry.get_service(model)
diff --git a/crudkit/integrations/fastapi.py b/crudkit/integrations/fastapi.py
new file mode 100644
index 0000000..c62f940
--- /dev/null
+++ b/crudkit/integrations/fastapi.py
@@ -0,0 +1,24 @@
+from __future__ import annotations
+from contextlib import contextmanager
+from fastapi import Depends
+from sqlalchemy.orm import Session
+from ..engines import CRUDKitRuntime
+
+_runtime = CRUDKitRuntime()
+
+@contextmanager
+def _session_scope():
+    SessionLocal = _runtime.session_factory
+    session: Session = SessionLocal()
+    try:
+        yield session
+        session.commit()
+    except Exception:
+        session.rollback()
+        raise
+    finally:
+        session.close()
+
+def get_db():
+    with _session_scope() as s:
+        yield s
diff --git a/crudkit/integrations/flask.py b/crudkit/integrations/flask.py
new file mode 100644
index 0000000..3cc319b
--- /dev/null
+++ b/crudkit/integrations/flask.py
@@ -0,0 +1,20 @@
+from __future__ import annotations
+from flask import Flask
+from sqlalchemy.orm import scoped_session
+from ..engines import CRUDKitRuntime
+from ..config import Config
+
+def init_app(app: Flask, *, runtime: CRUDKitRuntime | None = None, config: type[Config] | None == None):
+    """
+    Initializes CRUDKit for a Flask app. Provies `app.extensions['crudkit']`
+    with a runtime (engine + session_factory). Caller manages session lifecycle.
+    """
+    runtime = runtime or CRUDKitRuntime(config=config)
+    app.extensions.setdefault("crudkit", {})
+    app.extensions["crudkit"]["runtime"] = runtime
+
+    Session = runtime.session_factory
+    if Session is not None:
+        app.extensions["crudkit"]["Session"] = scoped_session(Session)
+
+    return runtime
diff --git a/crudkit/mixins.py b/crudkit/mixins.py
deleted file mode 100644
index 38a0e71..0000000
--- a/crudkit/mixins.py
+++ /dev/null
@@ -1,23 +0,0 @@
-import datetime as dt
-from sqlalchemy import Column, Integer, DateTime, Boolean
-from sqlalchemy.orm import declared_attr
-from sqlalchemy.ext.hybrid import hybrid_property
-
-class CrudMixin:
-    id = Column(Integer, primary_key=True)
-    created_at = Column(DateTime, default=dt.datetime.utcnow, nullable=False)
-    updated_at = Column(DateTime, default=dt.datetime.utcnow, onupdate=dt.datetime.utcnow, nullable=False)
-    deleted = Column("deleted", Boolean, default=False, nullable=False)
-    version = Column(Integer, default=1, nullable=False)
-
-    @hybrid_property
-    def is_deleted(self):
-        return self.deleted
-
-    def mark_deleted(self):
-        self.deleted = True
-        self.version += 1
-
-    @declared_attr
-    def __mapper_args__(cls):
-        return {"version_id_col": cls.version}
diff --git a/crudkit/registry.py b/crudkit/registry.py
new file mode 100644
index 0000000..fe3bfb7
--- /dev/null
+++ b/crudkit/registry.py
@@ -0,0 +1,120 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Any, Dict, Optional, Type, TypeVar, cast
+
+from flask import Flask
+from sqlalchemy.orm import Session
+
+from crudkit.core.service import CRUDService
+from crudkit.api.flask_api import generate_crud_blueprint
+from crudkit.engines import CRUDKitRuntime
+
+T = TypeVar("T")
+
+@dataclass
+class Registered:
+    model: Type[Any]
+    service: CRUDService[Any]
+    blueprint_name: str
+    url_prefix: str
+
+class CRUDRegistry:
+    """
+    Binds:
+        - name -> model class
+        - model class -> CRUDService (using CRUDKitRuntime.session_factory)
+        - model class -> Flask blueprint (via generate_crud_blueprint)
+    """
+    def __init__(self, runtime: CRUDKitRuntime):
+        self._rt = runtime
+        self._models_by_key: Dict[str, Type[Any]] = {}
+        self._services_by_model: Dict[Type[Any], CRUDService[Any]] = {}
+        self._bps_by_model: Dict[Type[Any], Registered] = {}
+
+    @staticmethod
+    def _key(model_or_name: Type[Any] | str) -> str:
+        return model_or_name.lower() if isinstance(model_or_name, str) else model_or_name.__name__.lower()
+
+    def get_model(self, key: str) -> Optional[Type[Any]]:
+        return self._models_by_key.get(key.lower())
+
+    def get_service(self, model: Type[T]) -> Optional[CRUDService[T]]:
+        return cast(Optional[CRUDService[T]], self._services_by_model.get(model))
+
+    def is_registered(self, model: Type[Any]) -> bool:
+        return model in self._services_by_model
+
+    def register_class(
+        self,
+        app: Flask,
+        model: Type[Any],
+        *,
+        url_prefix: Optional[str] = None,
+        blueprint_name: Optional[str] = None,
+        polymorphic: bool = False,
+        service_kwargs: Optional[dict] = None
+    ) -> Registered:
+        """
+        Register a model:
+            - store name -> class
+            - create a CRUDService bound to a Session from runtime.session_factory
+            - attach backend into from runtime.backend
+            - mount Flask blueprint at /api/<modelname> by default
+        Idempotent for each model.
+        """
+        key = self._key(model)
+        self._models_by_key.setdefault(key, model)
+
+        svc = self._services_by_model.get(model)
+        if svc is None:
+            SessionMaker = self._rt.session_factory
+            if SessionMaker is None:
+                raise RuntimeError("CRUDKitRuntime.session_factory is not initialized.")
+
+            svc = CRUDService(
+                model,
+                session_factory=SessionMaker,
+                polymorphic=polymorphic,
+                backend=self._rt.backend,
+                **(service_kwargs or {}),
+            )
+            self._services_by_model[model] = svc
+
+        reg = self._bps_by_model.get(model)
+        if reg:
+            return reg
+
+        prefix = url_prefix or f"/api/{key}"
+        bp_name = blueprint_name or f"crudkit.{key}"
+
+        bp = generate_crud_blueprint(model, svc)
+        bp.name = bp_name
+        app.register_blueprint(bp, url_prefix=prefix)
+
+        reg = Registered(model=model, service=svc, blueprint_name=bp_name, url_prefix=prefix)
+        self._bps_by_model[model] = reg
+        return reg
+
+    def register_many(
+        self,
+        app: Flask,
+        models: list[Type[Any]],
+        *,
+        base_prefix: str = "/api",
+        polymorphic: bool = False,
+        service_kwargs: Optional[dict] = None,
+    ) -> list[Registered]:
+        out: list[Registered] = []
+        for m in models:
+            key = self._key(m)
+            out.append(
+                self.register_class(
+                    app,
+                    m,
+                    url_prefix=f"{base_prefix}/{key}",
+                    polymorphic=polymorphic,
+                    service_kwargs=service_kwargs,
+                )
+            )
+        return out
diff --git a/crudkit/serialize.py b/crudkit/serialize.py
deleted file mode 100644
index 3ba6116..0000000
--- a/crudkit/serialize.py
+++ /dev/null
@@ -1,22 +0,0 @@
-def serialize(obj, *, fields=None, expand=None):
-    expand = set(expand or [])
-    fields = set(fields or [])
-    out = {}
-    # base columns
-    for col in obj.__table__.columns:
-        name = col.key
-        if fields and name not in fields:
-            continue
-        out[name] = getattr(obj, name)
-    # expansions
-    for rel in obj.__mapper__.relationships:
-        if rel.key not in expand:
-            continue
-        val = getattr(obj, rel.key)
-        if val is None:
-            out[rel.key] = None
-        elif rel.uselist:
-            out[rel.key] = [serialize(child) for child in val]
-        else:
-            out[rel.key] = serialize(val)
-    return out
\ No newline at end of file
diff --git a/crudkit/service.py b/crudkit/service.py
deleted file mode 100644
index a62ae18..0000000
--- a/crudkit/service.py
+++ /dev/null
@@ -1,169 +0,0 @@
-import sqlalchemy as sa
-from sqlalchemy import func, asc
-from sqlalchemy.exc import IntegrityError
-from sqlalchemy.orm import Session, aliased
-from sqlalchemy.inspection import inspect
-from sqlalchemy.sql.elements import UnaryExpression
-
-from .dsl import QuerySpec, build_query, split_sort_tokens
-from .eager import default_eager_policy
-
-def _dedup_order_by(ordering):
-    seen = set()
-    result = []
-    for ob in ordering:
-        col = ob.element if isinstance(ob, UnaryExpression) else ob
-        key = f"{col}-{getattr(ob, 'modifier', '')}-{getattr(ob, 'operator', '')}"
-        if key in seen:
-            continue
-        seen.add(key)
-        result.append(ob)
-    return result
-
-def _parse_sort_token(token: str):
-    token = token.strip()
-    direction = "asc"
-    if token.startswith('-'):
-        direction = "desc"
-        token = token[1:]
-    if ":" in token:
-        key, dirpart = token.rsplit(":", 1)
-        direction = "desc" if dirpart.lower().startswith("d") else "asc"
-        return key, direction
-    return token, direction
-
-def _apply_dotted_ordering(stmt, Model, sort_tokens):
-    """
-    stmt: a select(Model) statement
-    sort_tokens: list[str] like ["owner.identifier", "-brand.name"]
-    Returns: (stmt, alias_cache)
-    """
-    mapper = inspect(Model)
-    alias_cache = {}  # maps a path like "owner" or "brand" to its alias
-
-    for tok in sort_tokens:
-        path, direction = _parse_sort_token(tok)
-        parts = [p for p in path.split(".") if p]
-        if not parts:
-            continue
-
-        entity = Model
-        current_mapper = mapper
-        alias_path = []
-
-        # Walk relationships for all but the last part
-        for rel_name in parts[:-1]:
-            rel = current_mapper.relationships.get(rel_name)
-            if rel is None:
-                # invalid sort key; skip quietly or raise
-                # raise ValueError(f"Unknown relationship {current_mapper.class_.__name__}.{rel_name}")
-                entity = None
-                break
-
-            alias_path.append(rel_name)
-            key = ".".join(alias_path)
-
-            if key in alias_cache:
-                entity_alias = alias_cache[key]
-            else:
-                # build an alias and join
-                entity_alias = aliased(rel.mapper.class_)
-                stmt = stmt.outerjoin(entity_alias, getattr(entity, rel.key))
-                alias_cache[key] = entity_alias
-
-            entity = entity_alias
-            current_mapper = inspect(rel.mapper.class_)
-
-        if entity is None:
-            continue
-
-        col_name = parts[-1]
-        # Validate final column
-        if col_name not in current_mapper.columns:
-            # raise ValueError(f"Unknown column {current_mapper.class_.__name__}.{col_name}")
-            continue
-
-        col = getattr(entity, col_name) if entity is not Model else getattr(Model, col_name)
-        stmt = stmt.order_by(col.desc() if direction == "desc" else col.asc())
-
-    return stmt
-
-class CrudService:
-    def __init__(self, session: Session, eager_policy=default_eager_policy):
-        self.s = session
-        self.eager_policy = eager_policy
-
-    def create(self, Model, data, *, before=None, after=None):
-        if before: data = before(data) or data
-        obj = Model(**data)
-        self.s.add(obj)
-        self.s.flush()
-        if after: after(obj)
-        return obj
-
-    def get(self, Model, id, spec: QuerySpec | None = None):
-        spec = spec or QuerySpec()
-        stmt = build_query(Model, spec, self.eager_policy).where(Model.id == id)
-        return self.s.execute(stmt).scalars().first()
-
-    def list(self, Model, spec: QuerySpec):
-        stmt = build_query(Model, spec, self.eager_policy)
-
-        simple_sorts, dotted_sorts = split_sort_tokens(spec.order_by)
-        if dotted_sorts:
-            stmt = _apply_dotted_ordering(stmt, Model, dotted_sorts)
-
-        # count query
-        pk = getattr(Model, "id")  # adjust if not 'id'
-        count_base = stmt.with_only_columns(sa.distinct(pk)).order_by(None)
-        total = self.s.execute(
-            sa.select(sa.func.count()).select_from(count_base.subquery())
-        ).scalar_one()
-
-        if spec.page and spec.per_page:
-            offset = (spec.page - 1) * spec.per_page
-            stmt = stmt.limit(spec.per_page).offset(offset)
-
-        # ---- ORDER BY handling ----
-        mapper = inspect(Model)
-        pk_cols = mapper.primary_key
-
-        # Gather all clauses added so far
-        ordering = list(stmt._order_by_clauses)
-
-        # Append pk tie-breakers if not already present
-        existing_cols = {
-            str(ob.element if isinstance(ob, UnaryExpression) else ob)
-            for ob in ordering
-        }
-        for c in pk_cols:
-            if str(c) not in existing_cols:
-                ordering.append(asc(c))
-
-        # Dedup *before* applying
-        ordering = _dedup_order_by(ordering)
-
-        # Now wipe old order_bys and set once
-        stmt = stmt.order_by(None).order_by(*ordering)
-
-        rows = self.s.execute(stmt).scalars().all()
-        return rows, total
-
-    def update(self, obj, data, *, before=None, after=None):
-        if obj.is_deleted: raise ValueError("Cannot update a deleted record")
-        if before: data = before(obj, data) or data
-        for k, v in data.items(): setattr(obj, k, v)
-        obj.version += 1
-        if after: after(obj)
-        return obj
-
-    def soft_delete(self, obj, *, cascade=False, guard=None):
-        if guard and not guard(obj): raise ValueError("Delete blocked by guard")
-        # optionsl FK hygiene checks go here
-        obj.mark_deleted()
-        return obj
-
-    def undelete(self, obj):
-        obj.deleted = False
-        obj.version += 1
-        return obj
diff --git a/crudkit/ui/__init__.py b/crudkit/ui/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/crudkit/ui/fragments.py b/crudkit/ui/fragments.py
new file mode 100644
index 0000000..1f74f3a
--- /dev/null
+++ b/crudkit/ui/fragments.py
@@ -0,0 +1,839 @@
+import os
+import re
+
+from collections import OrderedDict
+from flask import current_app, url_for
+from jinja2 import Environment, FileSystemLoader, ChoiceLoader
+from sqlalchemy import inspect
+from sqlalchemy.orm import Load, RelationshipProperty, class_mapper, load_only, selectinload
+from sqlalchemy.orm.base import NO_VALUE
+from typing import Any, Dict, List, Optional, Tuple
+
+_ALLOWED_ATTRS = {
+    "class", "placeholder", "autocomplete", "inputmode", "pattern",
+    "min", "max", "step", "maxlength", "minlength",
+    "required", "readonly", "disabled",
+    "multiple", "size",
+    "id", "name", "value",
+}
+
+def get_env():
+    app = current_app
+    default_path = os.path.join(os.path.dirname(__file__), 'templates')
+    fallback_loader = FileSystemLoader(default_path)
+
+    return app.jinja_env.overlay(
+        loader=ChoiceLoader([app.jinja_loader, fallback_loader])
+    )
+
+def _normalize_rows_layout(layout: Optional[List[dict]]) -> Dict[str, dict]:
+    """
+    Create node dicts for each row and link parent->children.
+    Node shape:
+      {
+        'name': str,
+        'legend': Optional[str],
+        'attrs': dict,         # sanitized
+        'order': int,
+        'parent': Optional[str],
+        'children': list,      # list of node names (we'll expand later)
+        'fields': list,        # filled later
+      }
+    Always ensures a 'main' node exists.
+    """
+    nodes: Dict[str, dict] = {}
+
+    def make_node(name: str) -> dict:
+        node = nodes.get(name)
+        if node is None:
+            node = nodes[name] = {
+                "name": name,
+                "legend": None,
+                "attrs": {},
+                "order": 0,
+                "parent": None,
+                "children": [],
+                "fields": [],
+            }
+        return node
+
+    # seed nodes from layout
+    if isinstance(layout, list):
+        for item in layout:
+            name = item.get("name")
+            if not isinstance(name, str) or not name:
+                continue
+            node = make_node(name)
+            node["legend"] = item.get("legend")
+            node["attrs"] = _sanitize_attrs(item.get("attrs") or {})
+            try:
+                node["order"] = int(item.get("order") or 0)
+            except Exception:
+                node["order"] = 0
+            parent = item.get("parent")
+            node["parent"] = parent if isinstance(parent, str) and parent else None
+
+    # ensure main exists and is early-ordered
+    main = make_node("main")
+    if "order" not in main or main["order"] == 0:
+        main["order"] = -10
+
+    # default any unknown parents to main (except main itself)
+    for n in list(nodes.values()):
+        if n["name"] == "main":
+            n["parent"] = None
+            continue
+        p = n["parent"]
+        if p is None or p not in nodes or p == n["name"]:
+            n["parent"] = "main"
+
+    # detect cycles defensively; break by reparenting to main
+    visiting = set()
+    visited = set()
+
+    def visit(name: str):
+        if name in visited:
+            return
+        if name in visiting:
+            # cycle; break this node to main
+            nodes[name]["parent"] = "main"
+            return
+        visiting.add(name)
+        parent = nodes[name]["parent"]
+        if parent is not None:
+            visit(parent)
+        visiting.remove(name)
+        visited.add(name)
+
+    for nm in list(nodes.keys()):
+        visit(nm)
+
+    # compute children lists
+    for n in nodes.values():
+        n["children"] = []
+    for n in nodes.values():
+        p = n["parent"]
+        if p is not None:
+            nodes[p]["children"].append(n["name"])
+
+    # sort children by (order, name) for deterministic rendering
+    for n in nodes.values():
+        n["children"].sort(key=lambda nm: (nodes[nm]["order"], nodes[nm]["name"]))
+
+    return nodes
+
+def _assign_fields_to_rows(fields: List[dict], rows: Dict[str, dict]) -> List[dict]:
+    """
+    Put fields into their target row buckets by name (default 'main'),
+    then return a list of root nodes expanded with nested dicts ready for templates.
+    """
+    # assign fields
+    for f in fields:
+        row_name = f.get("row") or "main"
+        node = rows.get(row_name) or rows["main"]
+        node["fields"].append(f)
+
+    # expand tree into nested structures
+    def expand(name: str) -> dict:
+        n = rows[name]
+        return {
+            "name": n["name"],
+            "legend": n["legend"],
+            "attrs": n["attrs"],
+            "order": n["order"],
+            "fields": n["fields"],
+            "children": [expand(ch) for ch in n["children"]],
+        }
+
+    # roots are nodes with parent == None
+    roots = [expand(nm) for nm, n in rows.items() if n["parent"] is None]
+    roots.sort(key=lambda r: (r["order"], r["name"]))
+    return roots
+
+def _sanitize_attrs(attrs: Any) -> dict[str, Any]:
+    """
+    Whitelist attributes; allow data-* and aria-*; render True as boolean attr.
+    Drop False/None and anything not whitelisted.
+    """
+    if not isinstance(attrs, dict):
+        return {}
+    out: dict[str, Any] = {}
+    for k, v in attrs.items():
+        if not isinstance(k, str):
+            continue
+        elif isinstance(v, str):
+            if len(v) > 512:
+                v = v[:512]
+        if k.startswith("data-") or k.startswith("aria-") or k in _ALLOWED_ATTRS:
+            if isinstance(v, bool):
+                if v:
+                    out[k] = True
+            elif v is not None:
+                out[k] = str(v)
+
+    return out
+
+class _SafeObj:
+    """Attribute access that returns '' for missing/None instead of exploding."""
+    __slots__ = ("_obj",)
+    def __init__(self, obj): self._obj = obj
+    def __str__(self): return "" if self._obj is None else str(self._obj)
+    def __getattr__(self, name):
+        if self._obj is None:
+            return ""
+        val = getattr(self._obj, name, None)
+        if val is None:
+            return ""
+        return _SafeObj(val)
+
+def _coerce_fk_value(values: dict | None, instance: Any, base: str):
+    """
+    Resolve current selection for relationship 'base':
+      1) values['<base>_id']
+      2) values['<base>']['id'] or values['<base>'] if it's an int or numeric string
+      3) instance.<base> (if already loaded) -> use its .id   [safe for detached]
+      4) instance.<base>_id (if already loaded and not expired)
+    Never trigger a lazy load.
+    """
+    # 1) explicit *_id from values
+    if isinstance(values, dict):
+        key = f"{base}_id"
+        if key in values:
+            return values.get(key)
+
+        rel = values.get(base)
+        # 2a) nested dict with id
+        if isinstance(rel, dict):
+            vid = rel.get("id") or rel.get(key)
+            if vid is not None:
+                return vid
+        # 2b) scalar id
+        if isinstance(rel, int):
+            return rel
+        if isinstance(rel, str):
+            s = rel.strip()
+            if s.isdigit():
+                return s  # template compares as strings, so this is fine
+
+    # 3) use loaded relationship object (safe even if instance is detached)
+    if instance is not None:
+        try:
+            state = inspect(instance)
+            rel_attr = state.attrs.get(base)
+            if rel_attr is not None and rel_attr.loaded_value is not NO_VALUE:
+                rel_obj = rel_attr.loaded_value
+                if rel_obj is not None:
+                    rid = getattr(rel_obj, "id", None)
+                    if rid is not None:
+                        return rid
+            # 4) use loaded fk column if present and not expired
+            id_attr = state.attrs.get(f"{base}_id")
+            if id_attr is not None and id_attr.loaded_value is not NO_VALUE:
+                return id_attr.loaded_value
+        except Exception:
+            pass
+
+    return None
+
+def _is_many_to_one(mapper, name: str) -> Optional[RelationshipProperty]:
+    try:
+        prop = mapper.relationships[name]
+    except Exception:
+        return None
+    if isinstance(prop, RelationshipProperty) and prop.direction.name == 'MANYTOONE':
+        return prop
+    return None
+
+def _rel_for_id_name(mapper, name: str) -> tuple[Optional[str], Optional[RelationshipProperty]]:
+    if name.endswith("_id"):
+        base = name[:-3]
+        prop = _is_many_to_one(mapper, base)
+        return (base, prop) if prop else (None, None)
+    else:
+        prop = _is_many_to_one(mapper, name)
+        return (name, prop) if prop else (None, None)
+
+def _fk_options(session, related_model, label_spec):
+    simple_cols, rel_paths = _extract_label_requirements(label_spec)
+    q = session.query(related_model)
+
+    col_attrs = []
+    if hasattr(related_model, "id"):
+        col_attrs.append(getattr(related_model, "id"))
+    for name in simple_cols:
+        if hasattr(related_model, name):
+            col_attrs.append(getattr(related_model, name))
+    if col_attrs:
+        q = q.options(load_only(*col_attrs))
+
+    for rel_name, col_name in rel_paths:
+        rel_prop = getattr(related_model, rel_name, None)
+        if rel_prop is None:
+            continue
+        try:
+            target_cls = related_model.__mapper__.relationships[rel_name].mapper.class_
+            col_attr = getattr(target_cls, col_name, None)
+            if col_attr is None:
+                q = q.options(selectinload(rel_prop))
+            else:
+                q = q.options(selectinload(rel_prop).load_only(col_attr))
+        except Exception:
+            q = q.options(selectinload(rel_prop))
+
+    if simple_cols:
+        first = simple_cols[0]
+        if hasattr(related_model, first):
+            q = q.order_by(getattr(related_model, first))
+
+    rows = q.all()
+    return [
+        {
+            'value': getattr(opt, 'id'),
+            'label': _label_from_obj(opt, label_spec),
+        }
+        for opt in rows
+    ]
+
+def _normalize_field_spec(spec, mapper, session, label_specs_model_default):
+    """
+    Turn a user field spec into a concrete field dict the template understands.
+    """
+    name = spec['name']
+    base_rel_name, rel_prop = _rel_for_id_name(mapper, name)
+
+    field = {
+        "name": name if not base_rel_name else f"{base_rel_name}_id",
+        "label": spec.get("label", name),
+        "type": spec.get("type"),
+        "options": spec.get("options"),
+        "attrs": spec.get("attrs"),
+        "label_attrs": spec.get("label_attrs"),
+        "wrap": spec.get("wrap"),
+        "row": spec.get("row"),
+        "help": spec.get("help"),
+        "template": spec.get("template"),
+        "template_name": spec.get("template_name"),
+        "template_ctx": spec.get("template_ctx"),
+    }
+
+    if rel_prop:
+        if field["type"] is None:
+            field["type"] = "select"
+        if field["type"] == "select" and field.get("options") is None and session is not None:
+            related_model = rel_prop.mapper.class_
+            label_spec = (
+                spec.get("label_spec")
+                or label_specs_model_default.get(base_rel_name)
+                or getattr(related_model, "__crud_label__", None)
+                or "id"
+            )
+            field["options"] = _fk_options(session, related_model, label_spec)
+        return field
+
+    col = mapper.columns.get(name)
+    if field["type"] is None:
+        if col is not None and hasattr(col.type, "python_type"):
+            py = None
+            try:
+                py = col.type.python_type
+            except Exception:
+                pass
+            if py is bool:
+                field["type"] = "checkbox"
+            else:
+                field["type"] = "text"
+        else:
+            field["type"] = "text"
+
+    return field
+
+def _extract_label_requirements(spec: Any) -> tuple[list[str], list[tuple[str, str]]]:
+    """
+    From a label spec, return:
+      - simple_cols: ["name", "code"]
+      - rel_paths: [("room_function", "description"), ("owner", "last_name")]
+    """
+    simple_cols: list[str] = []
+    rel_paths: list[tuple[str, str]] = []
+
+    def ingest(token: str) -> None:
+        token = str(token).strip()
+        if not token:
+            return
+        if "." in token:
+            rel, col = token.split(".", 1)
+            if rel and col:
+                rel_paths.append((rel, col))
+        else:
+            simple_cols.append(token)
+
+    if spec is None or callable(spec):
+        return simple_cols, rel_paths
+
+    if isinstance(spec, (list, tuple)):
+        for a in spec:
+            ingest(a)
+        return simple_cols, rel_paths
+
+    if isinstance(spec, str):
+        # format string like "{first} {last}" or "{room_function.description} · {name}"
+        if "{" in spec and "}" in spec:
+            names = re.findall(r"{\s*([^}:\s]+)", spec)
+            for n in names:
+                ingest(n)
+        else:
+            ingest(spec)
+        return simple_cols, rel_paths
+
+    return simple_cols, rel_paths
+
+def _attrs_from_label_spec(spec: Any) -> list[str]:
+    """
+    Return a list of attribute names needed from the related model to compute the label.
+    Only simple attribute names are returned; dotted paths return just the first segment.
+    """
+    if spec is None:
+        return []
+    if callable(spec):
+        return []
+    if isinstance(spec, (list, tuple)):
+        return [str(a).split(".", 1)[0] for a in spec]
+    if isinstance(spec, str):
+        if "{" in spec and "}" in spec:
+            names = re.findall(r"{\s*([^}:\s]+)", spec)
+            return [n.split(".", 1)[0] for n in names]
+        return [spec.split(".", 1)[0]]
+    return []
+
+def _label_from_obj(obj: Any, spec: Any) -> str:
+    if spec is None:
+        for attr in ("label", "name", "title", "description"):
+            if hasattr(obj, attr):
+                val = getattr(obj, attr)
+                if not callable(val) and val is not None:
+                    return str(val)
+        if hasattr(obj, "id"):
+            return str(getattr(obj, "id"))
+        return object.__repr__(obj)
+
+    if isinstance(spec, (list, tuple)):
+        parts = []
+        for a in spec:
+            cur = obj
+            for part in str(a).split("."):
+                cur = getattr(cur, part, None)
+                if cur is None:
+                    break
+            parts.append("" if cur is None else str(cur))
+        return " ".join(p for p in parts if p)
+
+    if isinstance(spec, str) and "{" in spec and "}" in spec:
+        fields = re.findall(r"{\s*([^}:\s]+)", spec)
+        data: dict[str, Any] = {}
+        for f in fields:
+            root = f.split(".", 1)[0]
+            if root not in data:
+                val = getattr(obj, root, None)
+                data[root] = _SafeObj(val)
+
+        try:
+            return spec.format(**data)
+        except Exception:
+            return str(obj)
+
+    cur = obj
+    for part in str(spec).split("."):
+        cur = getattr(cur, part, None)
+        if cur is None:
+            return ""
+    return str(cur)
+
+def _val_from_row_or_obj(row: Dict[str, Any], obj: Any, dotted: str) -> Any:
+    """Best-effort deep get: try the projected row first, then the ORM object."""
+    val = _deep_get(row, dotted)
+    if val is None:
+        val = _deep_get_from_obj(obj, dotted)
+    return val
+
+def _matches_simple_condition(row: Dict[str, Any], obj: Any, cond: Dict[str, Any]) -> bool:
+    """
+    Supports:
+        {"field": "foo.bar", "eq": 10}
+        {"field": "foo", "ne": None}
+        {"field": "count", "gt": 0} (also lt, gte, lte)
+        {"field": "name", "in": ["a","b"]}
+        {"field": "thing", "is": None, | True | False}
+        {"any": [ ...subconds... ]} # OR
+        {"all": [ ...subconds... ]} # AND
+        {"not": { ...subcond... }}  # NOT
+    """
+    if "any" in cond:
+        return any(_matches_simple_condition(row, obj, c) for c in cond["any"])
+    if "all" in cond:
+        return all(_matches_simple_condition(row, obj, c) for c in cond["all"])
+    if "not" in cond:
+        return not _matches_simple_condition(row, obj, cond["not"])
+
+    field = cond.get("field")
+    val = _val_from_row_or_obj(row, obj, field) if field else None
+
+    if "is" in cond:
+        target = cond["is"]
+        if target is None:
+            return val is None
+        if isinstance(target, bool):
+            return bool(val) is target
+        return val is target
+
+    if "eq" in cond:
+        return val == cond["eq"]
+    if "ne" in cond:
+        return val != cond["ne"]
+    if "gt" in cond:
+        try: return val > cond["gt"]
+        except Exception: return False
+    if "lt" in cond:
+        try: return val < cond["lt"]
+        except Exception: return False
+    if "gte" in cond:
+        try: return val >= cond["gte"]
+        except Exception: return False
+    if "lte" in cond:
+        try: return val <= cond["lte"]
+        except Exception: return False
+    if "in" in cond:
+        try: return val in cond["in"]
+        except Exception: return False
+
+    return False
+
+def _row_class_for(row: Dict[str, Any], obj: Any, rules: Optional[List[Dict[str, Any]]]) -> Optional[str]:
+    """
+    rules is a list of:
+        {"when": <condition-dict>, "class": "table-warning fw-semibold"}
+    Multiple matching rules stack classes. Later wins on duplicates by normal CSS rules.
+    """
+    if not rules:
+        return None
+    classes = []
+    for rule in rules:
+        when = rule.get("when") or {}
+        if _matches_simple_condition(row, obj, when):
+            cls = rule.get("class")
+            if cls:
+                classes.append(cls)
+
+    return " ".join(dict.fromkeys(classes)) or None
+
+def _is_rel_loaded(obj, rel_name: str) -> bool:
+    try:
+        state = inspect(obj)
+        attr = state.attrs[rel_name]
+        return attr.loaded_value is not NO_VALUE
+    except Exception:
+        return False
+
+def _deep_get_from_obj(obj, dotted: str):
+    cur = obj
+    parts = dotted.split(".")
+    for i, part in enumerate(parts):
+        if i < len(parts) - 1 and not _is_rel_loaded(cur, part):
+            return None
+        cur = getattr(cur, part, None)
+        if cur is None:
+            return None
+    return cur
+
+def _deep_get(row: Dict[str, Any], dotted: str) -> Any:
+    if dotted in row:
+        return row[dotted]
+
+    cur = row
+    for part in dotted.split('.'):
+        if isinstance(cur, dict) and part in cur:
+            cur = cur[part]
+        else:
+            return None
+    return cur
+
+def _format_value(val: Any, fmt: Optional[str]) -> Any:
+    if fmt is None:
+        return val
+    try:
+        if fmt == "yesno":
+            return "Yes" if bool(val) else "No"
+        if fmt == "date":
+            return val.strftime("%Y-%m-%d") if hasattr(val, "strftime") else val
+        if fmt == "datetime":
+            return val.strftime("%Y-%m-%d %H:%M") if hasattr(val, "strftime") else val
+        if fmt == "time":
+            return val.strftime("%H:%M") if hasattr(val, "strftime") else val
+    except Exception:
+        return val
+    return val
+
+def _class_for(val: Any, classes: Optional[Dict[str, str]]) -> Optional[str]:
+    if not classes:
+        return None
+    key = "none" if val is None else str(val).lower()
+    return classes.get(key, classes.get("default"))
+
+def _build_href(spec: Dict[str, Any], row: Dict[str, Any], obj) -> Optional[str]:
+    if not spec:
+        return None
+    params = {}
+    for k, v in (spec.get("params") or {}).items():
+        if isinstance(v, str) and v.startswith("{") and v.endswith("}"):
+            key = v[1:-1]
+            val = _deep_get(row, key)
+            if val is None:
+                val = _deep_get_from_obj(obj, key)
+            params[k] = val
+        else:
+            params[k] = v
+    if any(v is None for v in params.values()):
+        return None
+    try:
+        return url_for('crudkit.' + spec["endpoint"], **params)
+    except Exception as e:
+        return None
+
+def _humanize(field: str) -> str:
+    return field.replace(".", " > ").replace("_", " ").title()
+
+def _normalize_columns(columns: Optional[List[Dict[str, Any]]], default_fields: List[str]) -> List[Dict[str, Any]]:
+    if not columns:
+        return [{"field": f, "label": _humanize(f)} for f in default_fields]
+
+    norm = []
+    for col in columns:
+        c = dict(col)
+        c.setdefault("label", _humanize(c["field"]))
+        norm.append(c)
+    return norm
+
+def _normalize_opts(opts: Dict[str, Any]) -> Dict[str, Any]:
+    """
+    Accept either:
+        render_table(..., object_class='user', row_classe[...])
+    or:
+        render_table(..., opts={'object_class': 'user', 'row_classes': [...]})
+
+    Returns a flat dict with top-level keys for convenience, while preserving
+    all original keys for the template.
+    """
+    if not isinstance(opts, dict):
+        return {}
+
+    flat = dict(opts)
+
+    nested = flat.get("opts")
+    if isinstance(nested, dict):
+        for k, v in nested.items():
+            flat.setdefault(k, v)
+
+    return flat
+
+def get_crudkit_template(env, name):
+    try:
+        return env.get_template(f'crudkit/{name}')
+    except Exception:
+        return env.get_template(name)
+
+def render_field(field, value):
+    env = get_env()
+
+    # 1) custom template field
+    field_type = field.get('type', 'text')
+    if field_type == 'template':
+        tname = field.get('template') or field.get('template_name')
+        if not tname:
+            return ""  # nothing to render
+        t = get_crudkit_template(env, tname)
+        # merge ctx with some sensible defaults
+        ctx = dict(field.get('template_ctx') or {})
+        # make sure templates always see these
+        ctx.setdefault('field', field)
+        ctx.setdefault('value', value)
+        return t.render(**ctx)
+
+    # 2) normal controls
+    template = get_crudkit_template(env, 'field.html')
+    return template.render(
+        field_name=field['name'],
+        field_label=field.get('label', field['name']),
+        value=value,
+        field_type=field_type,
+        options=field.get('options', None),
+        attrs=_sanitize_attrs(field.get('attrs') or {}),
+        label_attrs=_sanitize_attrs(field.get('label_attrs') or {}),
+        help=field.get('help'),
+    )
+
+
+def render_table(objects: List[Any], columns: Optional[List[Dict[str, Any]]] = None, **opts):
+    env = get_env()
+    template = get_crudkit_template(env, 'table.html')
+
+    if not objects:
+        return template.render(fields=[], rows=[])
+
+    flat_opts = _normalize_opts(opts)
+
+    proj = getattr(objects[0], "__crudkit_projection__", None)
+    row_dicts = [obj.as_dict(proj) for obj in objects]
+
+    default_fields = [k for k in row_dicts[0].keys() if k != "id"]
+    cols = _normalize_columns(columns, default_fields)
+
+    row_rules = (flat_opts.get("row_classes") or [])
+
+    disp_rows = []
+    for obj, rd in zip(objects, row_dicts):
+        cells = []
+        for col in cols:
+            field = col["field"]
+            raw = _deep_get(rd, field)
+            text = _format_value(raw, col.get("format"))
+            href = _build_href(col.get("link"), rd, obj) if col.get("link") else None
+            cls = _class_for(raw, col.get("classes"))
+            cells.append({"text": text, "href": href, "class": cls})
+
+        row_cls = _row_class_for(rd, obj, row_rules)
+        disp_rows.append({"id": rd.get("id"), "class": row_cls, "cells": cells})
+
+    return template.render(columns=cols, rows=disp_rows, kwargs=flat_opts)
+
+def render_form(
+    model_cls,
+    values,
+    session=None,
+    *,
+    fields_spec: Optional[list[dict]] = None,
+    label_specs: Optional[Dict[str, Any]] = None,
+    exclude: Optional[set[str]] = None,
+    overrides: Optional[Dict[str, Dict[str, Any]]] = None,
+    instance: Any = None,
+    layout: Optional[list[dict]] = None,
+    submit_attrs: Optional[dict[str, Any]] = None,
+    submit_label: Optional[str] = None,
+):
+    """
+    fields_spec: list of dicts describing fields in order. Each dict supports:
+        - name: "first_name" | "location" | "location_id" (required)
+        - label: override_label
+        - type: "text" | "textarea" | "checkbox" | "select" | "hidden" | ...
+        - label_spec: for relationship selects, e.g. "{name} - {room_function.description}"
+        - options: prebuilt list of {"value","label"}; skips querying if provided
+        - attrs: dict of arbitrary HTML attributes, e.g. {"required": True, "placeholder": "Jane"}
+        - help: small help text under the field
+    label_specs: legacy per-relationship label spec fallback ({"location": "..."}).
+    exclude: set of field names to hide.
+    overrides: legacy quick overrides keyed by field name (label/type/etc.)
+    instance: the ORM object backing the form; used to populate *_id values
+    layout: A list of dicts describing layouts for fields.
+    submit_attrs: A dict of attributes to apply to the submit button.
+    """
+    env = get_env()
+    template = get_crudkit_template(env, "form.html")
+    exclude = exclude or set()
+    overrides = overrides or {}
+    label_specs = label_specs or {}
+
+    mapper = class_mapper(model_cls)
+    fields: list[dict] = []
+    values_map = dict(values or {})  # we'll augment this with *_id selections
+
+    if fields_spec:
+        # Spec-driven path
+        for spec in fields_spec:
+            if spec["name"] in exclude:
+                continue
+            field = _normalize_field_spec(
+                {**spec, **overrides.get(spec["name"], {})},
+                mapper, session, label_specs
+            )
+            fields.append(field)
+
+        # After building fields, inject current values for any M2O selects
+        for f in fields:
+            name = f.get("name")
+            if isinstance(name, str) and name.endswith("_id"):
+                base = name[:-3]
+                rel_prop = mapper.relationships.get(base)
+                if isinstance(rel_prop, RelationshipProperty) and rel_prop.direction.name == "MANYTOONE":
+                    values_map[name] = _coerce_fk_value(values, instance, base)
+
+    else:
+        # Auto-generate path (your original behavior)
+        fk_fields = set()
+
+        # Relationships first
+        for prop in mapper.iterate_properties:
+            if isinstance(prop, RelationshipProperty) and prop.direction.name == 'MANYTOONE':
+                base = prop.key
+                if base in exclude or f"{base}_id" in exclude:
+                    continue
+                if session is None:
+                    continue
+
+                related_model = prop.mapper.class_
+                rel_label_spec = (
+                    label_specs.get(base)
+                    or getattr(related_model, "__crud_label__", None)
+                    or "id"
+                )
+                options = _fk_options(session, related_model, rel_label_spec)
+                base_field = {
+                    "name": f"{base}_id",
+                    "label": base,
+                    "type": "select",
+                    "options": options,
+                }
+                field = {**base_field, **overrides.get(f"{base}_id", {})}
+                fields.append(field)
+                fk_fields.add(f"{base}_id")
+
+                # NEW: set the current selection for this dropdown
+                values_map[f"{base}_id"] = _coerce_fk_value(values, instance, base)
+
+        # Then plain columns
+        for col in model_cls.__table__.columns:
+            if col.name in fk_fields or col.name in exclude:
+                continue
+            if col.name in ('id', 'created_at', 'updated_at'):
+                continue
+            if col.default or col.server_default or col.onupdate:
+                continue
+            base_field = {
+                "name": col.name,
+                "label": col.name,
+                "type": "checkbox" if getattr(col.type, "python_type", None) is bool else "text",
+            }
+            field = {**base_field, **overrides.get(col.name, {})}
+            if field.get("wrap"):
+                field["wrap"] = _sanitize_attrs(field["wrap"])
+            fields.append(field)
+
+        if submit_attrs:
+            submit_attrs = _sanitize_attrs(submit_attrs)
+
+        common_ctx = {"values": values_map, "instance": instance, "model_cls": model_cls, "session": session}
+        for f in fields:
+            if f.get("type") == "template":
+                base = dict(common_ctx)
+                base.update(f.get("template_ctx") or {})
+                f["template_ctx"] = base
+
+    # Build rows (supports nested layout with parents)
+    rows_map = _normalize_rows_layout(layout)
+    rows_tree = _assign_fields_to_rows(fields, rows_map)
+
+    return template.render(
+        rows=rows_tree,
+        fields=fields,  # keep for backward compatibility
+        values=values_map,
+        render_field=render_field,
+        submit_attrs=submit_attrs,
+        submit_label=submit_label
+    )
diff --git a/crudkit/ui/templates/field.html b/crudkit/ui/templates/field.html
new file mode 100644
index 0000000..0c0a39b
--- /dev/null
+++ b/crudkit/ui/templates/field.html
@@ -0,0 +1,59 @@
+{# show label unless hidden/custom #}
+{% if field_type != 'hidden' and field_label %}
+  <label for="{{ field_name }}"
+    {% if label_attrs %}{% for k,v in label_attrs.items() %}
+      {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+    {% endfor %}{% endif %}>
+    {{ field_label }}
+  </label>
+{% endif %}
+
+{% if field_type == 'select' %}
+  <select name="{{ field_name }}" id="{{ field_name }}"
+          {% if attrs %}{% for k,v in attrs.items() %}
+            {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+          {% endfor %}{% endif %}
+          {%- if not options %} disabled{% endif %}>
+    {% if options %}
+      <option value="">-- Select --</option>
+      {% for opt in options %}
+        <option value="{{ opt.value }}" {% if opt.value|string == value|string %}selected{% endif %}>
+          {{ opt.label }}
+        </option>
+      {% endfor %}
+    {% else %}
+      <option value="">-- No selection available --</option>
+    {% endif %}
+  </select>
+
+{% elif field_type == 'textarea' %}
+  <textarea name="{{ field_name }}" id="{{ field_name }}"
+            {% if attrs %}{% for k,v in attrs.items() %}
+              {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+            {% endfor %}{% endif %}>{{ value }}</textarea>
+
+{% elif field_type == 'checkbox' %}
+  <input type="checkbox" name="{{ field_name }}" id="{{ field_name }}" value="1"
+         {% if value %}checked{% endif %}
+         {% if attrs %}{% for k,v in attrs.items() %}
+           {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+         {% endfor %}{% endif %}>
+
+{% elif field_type == 'hidden' %}
+  <input type="hidden" name="{{ field_name }}" id="{{ field_name }}" value="{{ value }}">
+
+{% elif field_type == 'display' %}
+  <div {% if attrs %}{% for k,v in attrs.items() %}
+         {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+       {% endfor %}{% endif %}>{{ value }}</div>
+
+{% else %}
+  <input type="text" name="{{ field_name }}" id="{{ field_name }}" value="{{ value }}"
+         {% if attrs %}{% for k,v in attrs.items() %}
+           {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+         {% endfor %}{% endif %}>
+{% endif %}
+
+{% if help %}
+  <div class="form-text">{{ help }}</div>
+{% endif %}
diff --git a/crudkit/ui/templates/form.html b/crudkit/ui/templates/form.html
new file mode 100644
index 0000000..b073fc3
--- /dev/null
+++ b/crudkit/ui/templates/form.html
@@ -0,0 +1,40 @@
+<form method="POST">
+  {% macro render_row(row) %}
+    <!-- {{ row.name }} -->
+    {% if row.fields or row.children or row.legend %}
+      {% if row.legend %}<legend>{{ row.legend }}</legend>{% endif %}
+      <fieldset
+        {% if row.attrs %}{% for k,v in row.attrs.items() %}
+          {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+        {% endfor %}{% endif %}>
+        {% for field in row.fields %}
+          <div
+            {% if field.wrap %}{% for k,v in field.wrap.items() %}
+              {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+            {% endfor %}{% endif %}>
+            {{ render_field(field, values.get(field.name, '')) | safe }}
+          </div>
+        {% endfor %}
+        {% for child in row.children %}
+          {{ render_row(child) }}
+        {% endfor %}
+      </fieldset>
+    {% endif %}
+  {% endmacro %}
+
+  {% if rows %}
+    {% for row in rows %}
+      {{ render_row(row) }}
+    {% endfor %}
+  {% else %}
+    {% for field in fields %}
+      {{ render_field(field, values.get(field.name, '')) | safe }}
+    {% endfor %}
+  {% endif %}
+
+  <button type="submit"
+      {% if submit_attrs %}{% for k,v in submit_attrs.items() %}
+        {{k}}{% if v is not sameas true %}="{{ v }}"{% endif %}
+      {% endfor %}{% endif %}
+  >{{ submit_label if label else 'Save' }}</button>
+</form>
diff --git a/crudkit/ui/templates/table.html b/crudkit/ui/templates/table.html
new file mode 100644
index 0000000..d5f302f
--- /dev/null
+++ b/crudkit/ui/templates/table.html
@@ -0,0 +1,26 @@
+<table>
+  <thead>
+    <tr>
+      {% for col in columns %}
+        <th>{{ col.label }}</th>
+      {% endfor %}
+    </tr>
+  </thead>
+  <tbody>
+    {% if rows %}
+      {% for row in rows %}
+        <tr class="{{ row.class or '' }}">
+          {% for cell in row.cells %}
+            {% if cell.href %}
+              <td class="{{ cell.class or '' }}"><a href="{{ cell.href }}">{{ cell.text if cell.text is not none else '-' }}</a></td>
+            {% else %}
+              <td class="{{ cell.class or '' }}">{{ cell.text if cell.text is not none else '-' }}</td>
+            {% endif %}
+          {% endfor %}
+        </tr>
+      {% endfor %}
+    {% else %}
+      <tr><td colspan="{{ columns|length }}">No data.</td></tr>
+    {% endif %}
+  </tbody>
+</table>
diff --git a/example_app/app.py b/example_app/app.py
deleted file mode 100644
index 4061fcd..0000000
--- a/example_app/app.py
+++ /dev/null
@@ -1,27 +0,0 @@
-from flask import Flask, render_template
-from sqlalchemy import create_engine
-from sqlalchemy.orm import sessionmaker
-from .models import Base, Author, Book
-from crudkit.blueprint import make_blueprint as make_json_blueprint
-from crudkit.html import make_fragments_blueprint
-
-engine = create_engine("sqlite:///example.db", echo=True, future=True)
-SessionLocal = sessionmaker(bind=engine, expire_on_commit=False)
-
-def session_factory():
-    return SessionLocal()
-
-registry = {"author": Author, "book": Book}
-
-def create_app():
-    app = Flask(__name__)
-    Base.metadata.create_all(engine)
-    app.register_blueprint(make_json_blueprint(session_factory, registry), url_prefix="/api")
-    app.register_blueprint(make_fragments_blueprint(session_factory, registry), url_prefix="/ui")
-    @app.get("/demo")
-    def demo():
-        return render_template("demo.html")
-    return app
-
-if __name__ == "__main__":
-    create_app().run(debug=True)
diff --git a/example_app/models.py b/example_app/models.py
deleted file mode 100644
index 68ce56a..0000000
--- a/example_app/models.py
+++ /dev/null
@@ -1,18 +0,0 @@
-from typing import List
-from sqlalchemy import String, ForeignKey
-from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column,  relationship
-from crudkit import CrudMixin
-
-class Base(DeclarativeBase):
-    pass
-
-class Author(CrudMixin, Base):
-    __tablename__ = "author"
-    name: Mapped[str] = mapped_column(String(200), nullable=False)
-    books: Mapped[List["Book"]] = relationship(back_populates="author", cascade="all, delete-orphan")
-
-class Book(CrudMixin, Base):
-    __tablename__ = "book"
-    title: Mapped[str] = mapped_column(String(200), nullable=False)
-    author_id: Mapped[int] = mapped_column(ForeignKey("author.id"), nullable=False)
-    author: Mapped[Author] = relationship(back_populates="books")
diff --git a/example_app/seed.py b/example_app/seed.py
deleted file mode 100644
index 6422861..0000000
--- a/example_app/seed.py
+++ /dev/null
@@ -1,19 +0,0 @@
-from .app import SessionLocal, engine
-from .models import Base, Author, Book
-
-def run():
-    Base.metadata.create_all(engine)
-    s = SessionLocal()
-    a1 = Author(name="Ursula K. Le Guin")
-    a2 = Author(name="Octavia E. Butler")
-    s.add_all([
-        a1, a2,
-        Book(title="The Left Hand of Darkness", author=a1),
-        Book(title="A Wizard of Earthsea", author=a1),
-        Book(title="Parable of the Sower", author=a2),
-    ])
-    s.commit()
-    s.close()
-
-if __name__ == "__main__":
-    run()
diff --git a/example_app/templates/demo.html b/example_app/templates/demo.html
deleted file mode 100644
index 56491fb..0000000
--- a/example_app/templates/demo.html
+++ /dev/null
@@ -1,17 +0,0 @@
-<!-- templates/demo.html -->
-<!doctype html><meta charset="utf-8">
-<script src="https://unpkg.com/htmx.org@2.0.0"></script>
-<body>
-  <table class="table-auto w-full border">
-    <thead><tr><th class="px-3 py-2">ID</th><th class="px-3 py-2">Title</th><th class="px-3 py-2">Author</th><th></th></tr></thead>
-    <tbody id="rows"
-      hx-get="/ui/book/frag/rows?fields_csv=id,title,author.name&page=1&per_page=20"
-      hx-trigger="load" hx-target="this" hx-swap="innerHTML"></tbody>
-  </table>
-
-  <button hx-get="/ui/book/frag/form?hx=1&fields_csv=id,title,author.name"
-          hx-target="#modal-body" hx-swap="innerHTML"
-          onclick="document.getElementById('modal').showModal()">New Book</button>
-
-  <dialog id="modal"><div id="modal-body"></div></dialog>
-</body>
diff --git a/muck/app.py b/muck/app.py
new file mode 100644
index 0000000..ae3059a
--- /dev/null
+++ b/muck/app.py
@@ -0,0 +1,48 @@
+from flask import Flask
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker, scoped_session
+
+from crudkit import ProdConfig
+from crudkit.api.flask_api import generate_crud_blueprint
+from crudkit.core.service import CRUDService
+from crudkit.integrations.flask import init_app
+
+from muck.models.dbref import Base, Dbref
+from muck.models.exit import Exit
+from muck.models.player import Player
+from muck.models.program import Program
+from muck.models.room import Room
+from muck.models.thing import Thing
+from muck.init import bootstrap_world
+
+DATABASE_URL = "sqlite:///muck.db"
+
+engine = create_engine(DATABASE_URL, echo=True)
+SessionLocal = scoped_session(sessionmaker(bind=engine))
+
+Base.metadata.create_all(engine)
+
+session = SessionLocal()
+
+bootstrap_world(session)
+
+app = Flask(__name__)
+
+dbref_service = CRUDService(Dbref, session, polymorphic=True)
+exit_service = CRUDService(Exit, session)
+player_service = CRUDService(Player, session)
+program_service = CRUDService(Program, session)
+room_service = CRUDService(Room, session)
+thing_service = CRUDService(Thing, session)
+
+app.register_blueprint(generate_crud_blueprint(Dbref, dbref_service), url_prefix="/api/dbref")
+app.register_blueprint(generate_crud_blueprint(Exit, exit_service), url_prefix="/api/exits")
+app.register_blueprint(generate_crud_blueprint(Player, player_service), url_prefix="/api/players")
+app.register_blueprint(generate_crud_blueprint(Program, program_service), url_prefix="/api/programs")
+app.register_blueprint(generate_crud_blueprint(Room, room_service), url_prefix="/api/rooms")
+app.register_blueprint(generate_crud_blueprint(Thing, thing_service), url_prefix="/api/things")
+
+if __name__ == "__main__":
+    init_app(app, config=ProdConfig)
+    # app.run(debug=True, port=5050)
diff --git a/muck/init.py b/muck/init.py
new file mode 100644
index 0000000..03eca28
--- /dev/null
+++ b/muck/init.py
@@ -0,0 +1,39 @@
+from muck.models.room import Room
+from muck.models.player import Player
+
+def bootstrap_world(session):
+    if session.query(Room).first() or session.query(Player).first():
+        print("World already initialized.")
+        return
+
+    print("Bootstrapping world...")
+
+    room_zero = Room(
+        id=0,
+        name="Room Zero",
+        props={"_": {"de": "You are in Room Zero. It is very dark in here."}}
+    )
+
+    the_one = Player(
+        id=1,
+        name="One",
+        password="potrzebie",
+        props={"_": {"de": "You see The One."}}
+    )
+
+    the_one.location = room_zero
+    the_one.home = room_zero
+
+    the_one.creator = the_one
+    the_one.owner = the_one
+    the_one.modifier = the_one
+    the_one.last_user = the_one
+
+    room_zero.owner = the_one
+    room_zero.creator = the_one
+    room_zero.modifier = the_one
+    room_zero.last_user = the_one
+
+    session.add_all([room_zero, the_one])
+    session.commit()
+    print("World initialized.")
diff --git a/muck/models/__init__.py b/muck/models/__init__.py
new file mode 100644
index 0000000..1ee4e02
--- /dev/null
+++ b/muck/models/__init__.py
@@ -0,0 +1,5 @@
+from muck.models.dbref import Dbref
+from muck.models.exit import Exit
+from sqlalchemy.orm import relationship
+
+Dbref.exits = relationship("Exit", back_populates="source", foreign_keys=[Exit.location_id])
diff --git a/muck/models/dbref.py b/muck/models/dbref.py
new file mode 100644
index 0000000..5026161
--- /dev/null
+++ b/muck/models/dbref.py
@@ -0,0 +1,81 @@
+from sqlalchemy import Column, Integer, String, ForeignKey, Boolean, DateTime, JSON, Enum as SQLEnum, func
+from sqlalchemy.orm import relationship, foreign, remote
+from crudkit.core.base import CRUDMixin, Base
+from enum import Enum
+
+class ObjectType(str, Enum):
+    ROOM = "room"
+    THING = "thing"
+    EXIT = "exit"
+    PLAYER = "player"
+    PROGRAM = "program"
+
+TYPE_SUFFIXES = {
+    ObjectType.ROOM: "R",
+    ObjectType.EXIT: "E",
+    ObjectType.PLAYER: "P",
+    ObjectType.PROGRAM: "F",
+    ObjectType.THING: "",
+}
+
+class Dbref(Base, CRUDMixin):
+    __tablename__ = "dbref"
+
+    type = Column(SQLEnum(ObjectType, name="object_type_enum"), nullable=False)
+    name = Column(String, nullable=False)
+    props = Column(JSON, nullable=False, default={})
+    is_deleted = Column(Boolean, nullable=False, default=False)
+
+    last_used = Column(DateTime, nullable=False, default=func.now())
+    use_count = Column(Integer, nullable=False, default=0)
+
+    location_id = Column(Integer, ForeignKey("dbref.id"), nullable=True, default=0)
+    location = relationship("Dbref", foreign_keys=[location_id], back_populates="contents", primaryjoin=lambda: foreign(Dbref.location_id) == remote(Dbref.id), remote_side=lambda: Dbref.id)
+
+    contents = relationship("Dbref", foreign_keys=[location_id], back_populates="location")
+
+    owner_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    owner = relationship("Player", remote_side=[CRUDMixin.id], foreign_keys=[owner_id], primaryjoin=lambda: Dbref.owner_id == remote(Dbref.id), post_update=True)
+
+    creator_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    creator = relationship("Player", remote_side=[CRUDMixin.id], foreign_keys=[creator_id], primaryjoin=lambda: Dbref.creator_id == remote(Dbref.id), post_update=True)
+
+    modifier_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    modifier = relationship("Player", remote_side=[CRUDMixin.id], foreign_keys=[modifier_id], primaryjoin=lambda: Dbref.modifier_id == remote(Dbref.id), post_update=True)
+
+    last_user_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    last_user = relationship("Player", remote_side=[CRUDMixin.id], foreign_keys=[last_user_id], primaryjoin=lambda: Dbref.last_user_id == remote(Dbref.id), post_update=True)
+
+    __mapper_args__ = {
+        "polymorphic_on": type,
+        "polymorphic_identity": "dbref"
+    }
+
+    def __str__(self):
+        suffix = TYPE_SUFFIXES.get(self.type, "")
+        return f"#{self.id}{suffix}"
+
+    def __repr__(self):
+        suffix = TYPE_SUFFIXES.get(self.type, "")
+        return f"<Dbref #{self.id}{suffix}>"
+
+    def is_type(self, *types: ObjectType) -> bool:
+        return self.type in types
+
+    def display_type(self) -> str:
+        return self.type.value.upper()
+
+    @property
+    def is_room(self): return self.is_type(ObjectType.ROOM)
+
+    @property
+    def is_thing(self): return self.is_type(ObjectType.THING)
+
+    @property
+    def is_exit(self): return self.is_type(ObjectType.EXIT)
+
+    @property
+    def is_player(self): return self.is_type(ObjectType.PLAYER)
+
+    @property
+    def is_program(self): return self.is_type(ObjectType.PROGRAM)
diff --git a/muck/models/exit.py b/muck/models/exit.py
new file mode 100644
index 0000000..5523100
--- /dev/null
+++ b/muck/models/exit.py
@@ -0,0 +1,21 @@
+from sqlalchemy import Column, Integer, ForeignKey
+from sqlalchemy.orm import relationship, foreign, remote
+
+from crudkit.core.base import CRUDMixin
+
+from muck.models.dbref import Dbref, ObjectType
+
+class Exit(Dbref):
+    __tablename__ = "exits"
+
+    id = Column(Integer, ForeignKey("dbref.id"), primary_key=True)
+
+    destination_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    destination = relationship("Dbref", remote_side=[CRUDMixin.id], foreign_keys=[destination_id], primaryjoin=lambda: foreign(Exit.destination_id) == remote(Dbref.id))
+
+    source = relationship("Dbref", back_populates="exits", foreign_keys=[Dbref.location_id], remote_side=[Dbref.id])
+
+    __mapper_args__ = {
+        "polymorphic_identity": ObjectType.EXIT,
+        "inherit_condition": id == Dbref.id
+    }
diff --git a/muck/models/player.py b/muck/models/player.py
new file mode 100644
index 0000000..6e1db75
--- /dev/null
+++ b/muck/models/player.py
@@ -0,0 +1,27 @@
+from sqlalchemy import Column, Integer, Boolean, String, ForeignKey
+from sqlalchemy.orm import relationship, foreign, remote
+
+from crudkit.core.base import CRUDMixin
+
+from muck.models.dbref import Dbref, ObjectType
+
+class Player(Dbref):
+    __tablename__ = "players"
+
+    id = Column(Integer, ForeignKey("dbref.id"), primary_key=True)
+
+    pennies = Column(Integer, nullable=False, default=0)
+    insert_mode = Column(Boolean, nullable=False, default=False)
+    block = Column(Integer, nullable=True)
+    password = Column(String, nullable=False)
+
+    home_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    home = relationship("Dbref", remote_side=[CRUDMixin.id], foreign_keys=[home_id], primaryjoin=lambda: foreign(Player.home_id) == remote(Dbref.id))
+
+    current_program_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    current_program = relationship("Dbref", remote_side=[CRUDMixin.id], foreign_keys=[current_program_id], primaryjoin=lambda: foreign(Player.current_program_id) == remote(Dbref.id))
+
+    __mapper_args__ = {
+        "polymorphic_identity": ObjectType.PLAYER,
+        "inherit_condition": id == Dbref.id
+    }
diff --git a/muck/models/program.py b/muck/models/program.py
new file mode 100644
index 0000000..14548cb
--- /dev/null
+++ b/muck/models/program.py
@@ -0,0 +1,13 @@
+from sqlalchemy import Column, Integer, ForeignKey
+
+from muck.models.dbref import Dbref, ObjectType
+
+class Program(Dbref):
+    __tablename__ = "programs"
+
+    id = Column(Integer, ForeignKey("dbref.id"), primary_key=True)
+
+    __mapper_args__ = {
+        "polymorphic_identity": ObjectType.PROGRAM,
+        "inherit_condition": id == Dbref.id
+    }
diff --git a/muck/models/room.py b/muck/models/room.py
new file mode 100644
index 0000000..b0e5695
--- /dev/null
+++ b/muck/models/room.py
@@ -0,0 +1,19 @@
+from sqlalchemy import Column, Integer, ForeignKey
+from sqlalchemy.orm import relationship, foreign, remote
+
+from crudkit.core.base import CRUDMixin
+
+from muck.models.dbref import Dbref, ObjectType
+
+class Room(Dbref):
+    __tablename__ = "rooms"
+
+    id = Column(Integer, ForeignKey("dbref.id"), primary_key=True)
+
+    dropto_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    dropto = relationship("Dbref", remote_side=[CRUDMixin.id], foreign_keys=[dropto_id], primaryjoin=lambda: foreign(Room.dropto_id) == remote(Dbref.id))
+
+    __mapper_args__ = {
+        "polymorphic_identity": ObjectType.ROOM,
+        "inherit_condition": id == Dbref.id
+    }
diff --git a/muck/models/thing.py b/muck/models/thing.py
new file mode 100644
index 0000000..c67bc27
--- /dev/null
+++ b/muck/models/thing.py
@@ -0,0 +1,21 @@
+from sqlalchemy import Column, Integer, ForeignKey
+from sqlalchemy.orm import relationship, foreign, remote
+
+from crudkit.core.base import CRUDMixin
+
+from muck.models.dbref import Dbref, ObjectType
+
+class Thing(Dbref):
+    __tablename__ = "things"
+
+    id = Column(Integer, ForeignKey("dbref.id"), primary_key=True)
+
+    value = Column(Integer, nullable=False, default=0)
+
+    home_id = Column(Integer, ForeignKey("dbref.id"), nullable=True)
+    home = relationship("Dbref", remote_side=[CRUDMixin.id], foreign_keys=[home_id], primaryjoin=lambda: foreign(Thing.home_id) == remote(Dbref.id))
+
+    __mapper_args__ = {
+        "polymorphic_identity": ObjectType.THING,
+        "inherit_condition": id == Dbref.id
+    }