fix(history): defer full transcript hydration to model sends (#5929)

* fix(history): defer full hydration to model sends

* fix(session): key hydration on real rows, fork through get_session

Two regressions from the display/model-context split, both reproducible
against dev.

The hydration gate compared the cached transcript against the
denormalized sessions.message_count column. That column drifts in normal
operation — _persist_message swallows a failed insert while add_message
has already appended in memory, so the next successful persist writes
rows+1 — and _db_to_session re-read the same column after each reload, so
the shortfall never closed. Every send, edit, delete and truncate on a
warm session re-selected the whole message table: the cost this change
set out to remove, relocated onto the hot path. The other direction was
just as bad — a persist for an uncached session writes message_count = 0,
and a stale-low counter with a partly filled cache meant no hydration at
all and a silently truncated transcript for the model.

sync_session_metadata now reconciles message_count against COUNT(*) on
chat_messages (one indexed count inside the connection it already opens),
and _db_to_session trusts the rows it just loaded. A hydrate always
closes the gap, so the next read is a cache hit.

fork_session read session_manager.sessions directly and never hydrated.
keep_count indexes into source.history, and display pagination no longer
fills that cache, so forking after a restart returned HTTP 200 with an
empty conversation and no error surfaced. It goes through get_session
now.

_hydrate_session_history_from_db is gone with its helper: get_session is
the hydration seam, and rebuilding session.history from raw rows in the
display fallback overwrote the parsed multimodal content and the _db_id
edit/delete keys that had just been set.

Tests drive a real SessionManager over a temp DB instead of a stub that
only proved the stub hydrates — both drift directions, the send path
warm and cold, and a fork taken after a restart. All five fail without
this change. The brittle SQL-text assertions are dropped; the page
bounds are already proven by the response body.

* fix(history): route pagination through canonical handler

---------

Co-authored-by: Léo <leograndcontact@gmail.com>

core/session_manager.py

@@ -194,7 +194,12 @@ class SessionManager:
             is_important=getattr(db_session, 'is_important', False) or False,
         )
 
-        session.message_count = getattr(db_session, 'message_count', len(history))
+        # The rows just loaded are the whole transcript, so they — not the
+        # denormalized sessions.message_count column — are the truth for this
+        # cached object. get_session's hydration gate compares against this
+        # number; seeding it from a drifted column would ask for a reload that
+        # can never close the gap.
+        session.message_count = len(history)
         return session
 
     # ------------------------------------------------------------------
@@ -398,30 +403,50 @@ class SessionManager:
     # ------------------------------------------------------------------
 
     def get_session(self, session_id: str) -> Session:
-        """Get a session by ID, loading from DB if needed.
+        """Get a session by ID, loading complete DB history when needed.
 
-        Sessions seeded by `load_sessions` start with empty history. The
+        Sessions seeded by ``load_sessions`` start with empty history, and a
-        first read here hydrates them with the message rows.
+        cached session can also become partially stale. Refresh metadata first,
+        then hydrate whenever the cached transcript is short of the stored rows.
+        Model-send routes enter through this method before building context,
+        while paginated display history reads SQLite directly.
+
+        The gate compares against ``sync_session_metadata``'s reconciled count
+        (the real ``chat_messages`` total), never the denormalized column, so a
+        hydrate always closes the gap and the next read is a cache hit.
         """
         if session_id not in self.sessions:
             self._load_session_from_db(session_id)
-        else:
-            cached = self.sessions[session_id]
-            # Lazy hydrate: metadata-only entries get their messages on first read.
-            if not cached.history and getattr(cached, "message_count", 0) > 0:
-                self._load_session_from_db(session_id)
 
         # Keep model/endpoint metadata fresh. Endpoint deletion can clear the
-        # DB row while a session object is still cached in RAM.
+        # DB row while a session object is still cached in RAM. Refreshing first
+        # also exposes the authoritative message count before completeness is
+        # checked.
         self.sync_session_metadata(session_id)
 
+        cached = self.sessions[session_id]
+        cached_count = len(cached.history or [])
+        stored_count = int(getattr(cached, "message_count", 0) or 0)
+        if cached_count < stored_count:
+            self._load_session_from_db(session_id)
+
         # Update last_accessed
         self._touch_session(session_id)
 
         return self.sessions[session_id]
 
     def sync_session_metadata(self, session_id: str) -> bool:
-        """Refresh non-message session fields from the DB into the cached object."""
+        """Refresh non-message session fields from the DB into the cached object.
+
+        ``message_count`` is reconciled against the real ``chat_messages`` rows
+        rather than copied from the denormalized ``sessions.message_count``
+        column. That column drifts in normal operation — ``_persist_message``
+        swallows a failed insert but ``add_message`` has already appended in
+        memory, so the next successful persist writes rows+1, and a persist for
+        an uncached session writes 0. Hydration keys off this number: a
+        drifted-high column would reload the whole transcript on every warm
+        read, and a drifted-low one would leave the model a truncated one.
+        """
         session = self.sessions.get(session_id)
         if session is None:
             return False
@@ -444,7 +469,11 @@ class SessionManager:
             session.archived = db_session.archived
             session.owner = getattr(db_session, "owner", None)
             session.is_important = getattr(db_session, "is_important", False) or False
-            session.message_count = getattr(db_session, "message_count", session.message_count) or 0
+            session.message_count = (
+                db.query(DbChatMessage)
+                .filter(DbChatMessage.session_id == session_id)
+                .count()
+            )
             return True
         except Exception as e:
             logger.error(f"Error syncing session metadata {session_id}: {e}")

routes/history/history_routes.py

@@ -137,44 +137,6 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
             entry["metadata"] = meta
         return entry
 
-    def _db_message_metadata(m: DbChatMessage) -> Dict[str, Any]:
-        meta = {}
-        if m.meta_data:
-            try:
-                meta = json.loads(m.meta_data) or {}
-            except (json.JSONDecodeError, ValueError):
-                meta = {}
-        if m.timestamp and "timestamp" not in meta:
-            meta["timestamp"] = m.timestamp.isoformat() + "Z"
-        return meta
-
-    def _hydrate_session_history_from_db(session_id: str, rows: list[DbChatMessage]) -> None:
-        """Rebuild in-memory context from raw DB rows after a history load.
-
-        The browser history endpoint can return paged/display-trimmed messages,
-        but the next model call reads ``session.history``. After a restart or a
-        stale in-memory session, selecting an old chat through the paged endpoint
-        used to show the transcript while the model only saw fresh context.
-        """
-        if not rows:
-            return
-        try:
-            session = session_manager.get_session(session_id)
-        except KeyError:
-            return
-        session.history = [
-            ChatMessage(role=m.role, content=m.content, metadata=_db_message_metadata(m) or None)
-            for m in rows
-        ]
-        session.message_count = len(session.history)
-
-    def _session_needs_db_history_hydration(session_id: str, total: int) -> bool:
-        try:
-            session = session_manager.get_session(session_id)
-        except KeyError:
-            return False
-        return len(session.history or []) < int(total or 0)
-
     @router.get("/api/history/{session_id}")
     async def get_session_history(
         request: Request,
@@ -198,6 +160,8 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
                 )
                 page_offset = int(offset) if offset is not None else max(total - page_limit, 0)
                 page_offset = max(0, min(page_offset, total))
+                # Keep display pagination page-scoped. ``get_session`` is the
+                # full model-context hydration seam and must not be entered here.
                 rows = (
                     db.query(DbChatMessage)
                     .filter(DbChatMessage.session_id == session_id)
@@ -206,14 +170,6 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
                     .limit(page_limit)
                     .all()
                 )
-                if _session_needs_db_history_hydration(session_id, total):
-                    full_rows = (
-                        db.query(DbChatMessage)
-                        .filter(DbChatMessage.session_id == session_id)
-                        .order_by(DbChatMessage.timestamp)
-                        .all()
-                    )
-                    _hydrate_session_history_from_db(session_id, full_rows)
                 history_dict = [
                     entry for entry in (_db_history_entry(m) for m in rows)
                     if not (entry.get("metadata") or {}).get("hidden")
@@ -258,7 +214,10 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
                     entry["metadata"] = msg["metadata"]
                 history_dict.append(entry)
 
-        # Fallback: load from DB if in-memory is empty
+        # Fallback: load from DB if in-memory renders empty. Display only —
+        # get_session above is the hydration seam, so nothing here writes back
+        # into session.history — rebuilding it from raw rows would overwrite
+        # parsed multimodal content and the _db_id edit/delete keys it just set.
         if not history_dict:
             db = SessionLocal()
             try:
@@ -268,17 +227,10 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
                     .order_by(DbChatMessage.timestamp)
                     .all()
                 )
-                db_history = []
-                for m in db_messages:
-                    db_history.append(_db_history_entry(m))
-                if db_history:
-                    # Rebuild in-memory history from the full set so hidden
-                    # messages (e.g. compaction summaries) are kept for AI context.
-                    _hydrate_session_history_from_db(session_id, db_messages)
                 # Response excludes hidden messages, matching the in-memory path.
                 history_dict = [
-                    m for m in db_history
+                    entry for entry in (_db_history_entry(m) for m in db_messages)
-                    if not (m.get("metadata") or {}).get("hidden")
+                    if not (entry.get("metadata") or {}).get("hidden")
                 ]
             except Exception as e:
                 logger.error(f"DB fallback failed for {session_id}: {e}")
@@ -645,8 +597,14 @@ def setup_history_routes(session_manager, upload_handler=None) -> APIRouter:
             body = await request.json()
             keep_count = body.get("keep_count", 0)
 
-            # Get the source session
+            # Get the source session. keep_count indexes into source.history,
-            source = session_manager.sessions.get(session_id)
+            # so this must go through get_session — reading the cache directly
+            # forks an empty transcript out of a metadata-only session after a
+            # restart (display pagination no longer hydrates it).
+            try:
+                source = session_manager.get_session(session_id)
+            except KeyError:
+                raise HTTPException(404, "Session not found")
             if not source:
                 raise HTTPException(404, "Session not found")
 

routes/session_routes.py

@@ -801,15 +801,6 @@ def setup_session_routes(
         finally:
             db.close()
 
-    @router.get("/history/{sid}")
-    def get_history(request: Request, sid: str):
-        _verify_session_owner(request, sid)
-        try:
-            session = session_manager.get_session(sid)
-        except KeyError:
-            raise HTTPException(404, f"Session {sid} not found")
-        return {"history": [msg.to_dict() for msg in session.history]}
-    
     @router.get("/session/{sid}/export")
     def export_session(request: Request, sid: str, fmt: str = "md", filename: str = ""):
         """Export conversation history as a downloadable file.

tests/test_fork_session_metadata.py

@@ -34,6 +34,11 @@ class _FakeSessionManager:
         self.sessions = {"src-id": source}
         self.created = None
 
+    def get_session(self, session_id):
+        # Fork looks the source up through get_session — the hydration seam —
+        # so a session only present in the DB still forks a real transcript.
+        return self.sessions[session_id]
+
     def create_session(self, session_id=None, name=None, endpoint_url=None,
                        model=None, rag=False, owner=None):
         self.created = _FakeSession(name=name, owner=owner)

tests/test_history_db_fallback_hidden.py

@@ -5,9 +5,9 @@ The in-memory branch skips messages whose metadata has ``hidden`` (e.g.
 compaction summaries that are kept for AI context but not shown to the user).
 The DB fallback (taken when the in-memory history is empty, e.g. after a
 restart) built the client response from every DB row with no such filter, so
-hidden messages leaked to the client on DB-served sessions. The rebuilt
+hidden messages leaked to the client on DB-served sessions. Hydration of
-in-memory ``session.history`` must still keep them, though, so only the response
+``session.history`` belongs to ``get_session``; this fallback only shapes the
-is filtered.
+response, so only the response is filtered.
 
 get_session_history depends on the DB, the session manager and a FastAPI
 request, so this pins the regression at the source level (as other route tests

tests/test_history_display_model_hydration.py

@@ -0,0 +1,549 @@
+"""Display pagination must stay separate from full model-context hydration."""
+
+import json
+from datetime import datetime, timedelta
+
+import pytest
+from fastapi import APIRouter, FastAPI
+from fastapi.testclient import TestClient
+from starlette.requests import Request
+from sqlalchemy import create_engine, event
+from sqlalchemy.orm import sessionmaker
+from sqlalchemy.pool import StaticPool
+
+from core.database import Base, ChatMessage as DbChatMessage, Session as DbSession
+from core.models import ChatMessage, Session
+from core.session_manager import SessionManager
+from routes import chat_routes
+from routes.history import history_routes
+from routes import session_routes
+from src.request_models import ChatRequest
+
+
+def _database():
+    engine = create_engine(
+        "sqlite://",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(
+        engine,
+        tables=[DbSession.__table__, DbChatMessage.__table__],
+    )
+    return engine, sessionmaker(bind=engine, autocommit=False, autoflush=False)
+
+
+def _seed_session(db_factory, *, session_id="session-1", message_count=6, stored_count=None):
+    """Seed `message_count` real rows; `stored_count` overrides the denormalized
+    sessions.message_count column so drift can be reproduced."""
+    db = db_factory()
+    try:
+        db.add(
+            DbSession(
+                id=session_id,
+                name="Long chat",
+                endpoint_url="http://model.test/v1",
+                model="test-model",
+                owner="alice",
+                message_count=message_count if stored_count is None else stored_count,
+            )
+        )
+        start = datetime(2026, 1, 1, 12, 0, 0)
+        for index in range(message_count):
+            db.add(
+                DbChatMessage(
+                    id=f"message-{index}",
+                    session_id=session_id,
+                    role="user" if index % 2 == 0 else "assistant",
+                    content=f"content-{index}",
+                    timestamp=start + timedelta(seconds=index),
+                )
+            )
+        db.commit()
+    finally:
+        db.close()
+
+
+def _chat_message_selects(statements):
+    return [
+        " ".join(statement.lower().split())
+        for statement in statements
+        if statement.lstrip().lower().startswith("select")
+        and "chat_messages" in statement.lower()
+    ]
+
+
+def _manager(db_factory, monkeypatch, sessions=None):
+    """A real SessionManager bound to the temp DB, with load counting."""
+    monkeypatch.setattr("core.session_manager.SessionLocal", db_factory)
+    manager = object.__new__(SessionManager)
+    manager.upload_handler = None
+    manager.sessions = sessions if sessions is not None else {}
+    manager.full_loads = 0
+
+    original_load = manager._load_session_from_db
+
+    def counting_load(session_id):
+        manager.full_loads += 1
+        return original_load(session_id)
+
+    manager._load_session_from_db = counting_load
+    return manager
+
+
+def test_paginated_history_reads_only_count_and_requested_page(monkeypatch):
+    engine, db_factory = _database()
+    _seed_session(db_factory)
+
+    class DisplayOnlyManager:
+        def get_session(self, _session_id):
+            raise AssertionError("paginated display history must not hydrate model context")
+
+    monkeypatch.setattr(history_routes, "SessionLocal", db_factory)
+    monkeypatch.setattr(history_routes, "_verify_session_owner", lambda *_args: None)
+
+    app = FastAPI()
+    app.include_router(history_routes.setup_history_routes(DisplayOnlyManager()))
+
+    statements = []
+
+    def capture_sql(_conn, _cursor, statement, _parameters, _context, _executemany):
+        statements.append(statement)
+
+    event.listen(engine, "before_cursor_execute", capture_sql)
+    try:
+        response = TestClient(app).get("/api/history/session-1?limit=2")
+    finally:
+        event.remove(engine, "before_cursor_execute", capture_sql)
+        engine.dispose()
+
+    assert response.status_code == 200
+    payload = response.json()
+    assert [message["content"] for message in payload["history"]] == [
+        "content-4",
+        "content-5",
+    ]
+    assert payload["total"] == 6
+    assert payload["offset"] == 4
+    assert payload["has_more_before"] is True
+    assert payload["has_more_after"] is False
+
+    # One COUNT for the total plus one page read — never a full-transcript
+    # select. The page bounds are asserted through the response above rather
+    # than by matching SQL text.
+    chat_selects = _chat_message_selects(statements)
+    assert len(chat_selects) == 2, chat_selects
+    assert sum("count(" in statement for statement in chat_selects) == 1
+
+
+def test_production_router_order_reaches_bounded_canonical_history(monkeypatch):
+    """The assembled app must not shadow canonical history with session routes."""
+    engine, db_factory = _database()
+    _seed_session(db_factory, message_count=1200)
+
+    class DisplayOnlyManager:
+        def get_session(self, _session_id):
+            raise AssertionError("bounded initial history must not hydrate all messages")
+
+    manager = DisplayOnlyManager()
+    monkeypatch.setattr(
+        session_routes,
+        "router",
+        APIRouter(prefix="/api", tags=["sessions"]),
+    )
+    monkeypatch.setattr(history_routes, "SessionLocal", db_factory)
+    monkeypatch.setattr(history_routes, "_verify_session_owner", lambda *_args: None)
+
+    app = FastAPI()
+    app.include_router(session_routes.setup_session_routes(manager, {}))
+    app.include_router(history_routes.setup_history_routes(manager))
+
+    try:
+        response = TestClient(app).get("/api/history/session-1?limit=24")
+    finally:
+        engine.dispose()
+
+    assert response.status_code == 200
+    assert response.request.url.params["limit"] == "24"
+    payload = response.json()
+    displayed = len(payload["history"])
+    assert 0 < displayed <= payload["limit"] <= 100
+    assert payload["total"] >= 1200
+    assert payload["has_more_before"] is True
+    assert displayed < payload["total"]
+
+
+def test_incomplete_cached_history_hydrates_once_for_model_context(monkeypatch):
+    engine, db_factory = _database()
+    raw_multimodal = json.dumps(
+        [
+            {"type": "text", "text": "look at the source image"},
+            {
+                "type": "image_url",
+                "image_url": {"url": "data:image/png;base64,AAAA"},
+            },
+        ]
+    )
+    db = db_factory()
+    try:
+        db.add(
+            DbSession(
+                id="session-1",
+                name="Long chat",
+                endpoint_url="http://model.test/v1",
+                model="test-model",
+                owner="alice",
+                message_count=3,
+            )
+        )
+        start = datetime(2026, 1, 1, 12, 0, 0)
+        db.add_all(
+            [
+                DbChatMessage(
+                    id="message-0",
+                    session_id="session-1",
+                    role="user",
+                    content=raw_multimodal,
+                    meta_data=json.dumps(
+                        {
+                            "attachments": [
+                                {
+                                    "id": "upload-1",
+                                    "filename": "source.png",
+                                    "content_type": "image/png",
+                                }
+                            ]
+                        }
+                    ),
+                    timestamp=start,
+                ),
+                DbChatMessage(
+                    id="message-1",
+                    session_id="session-1",
+                    role="assistant",
+                    content="answer",
+                    timestamp=start + timedelta(seconds=1),
+                ),
+                DbChatMessage(
+                    id="message-2",
+                    session_id="session-1",
+                    role="system",
+                    content="compaction summary",
+                    meta_data=json.dumps({"hidden": True}),
+                    timestamp=start + timedelta(seconds=2),
+                ),
+            ]
+        )
+        db.commit()
+    finally:
+        db.close()
+
+    manager = _manager(
+        db_factory,
+        monkeypatch,
+        sessions={
+            "session-1": Session(
+                id="session-1",
+                name="Long chat",
+                endpoint_url="http://model.test/v1",
+                model="test-model",
+                owner="alice",
+                history=[ChatMessage("user", "stale partial cache")],
+                # Deliberately stale too: get_session must refresh metadata before
+                # checking whether the cached transcript is complete.
+                message_count=1,
+            )
+        },
+    )
+
+    try:
+        hydrated = manager.get_session("session-1")
+        first_full_loads = manager.full_loads
+        warm = manager.get_session("session-1")
+        second_full_loads = manager.full_loads
+    finally:
+        engine.dispose()
+
+    assert hydrated is warm
+    assert len(hydrated.history) == 3
+    assert first_full_loads == 1
+    assert second_full_loads == first_full_loads
+
+    context = hydrated.get_context_messages()
+    assert len(context) == 3
+    assert context[0]["content"][1]["image_url"]["url"] == "data:image/png;base64,AAAA"
+    assert context[0]["metadata"]["attachments"] == [
+        {
+            "id": "upload-1",
+            "filename": "source.png",
+            "content_type": "image/png",
+        }
+    ]
+    hidden_summary = next(message for message in context if message["role"] == "system")
+    assert hidden_summary["content"] == "compaction summary"
+    assert hidden_summary["metadata"]["hidden"] is True
+
+
+def test_inflated_message_count_column_does_not_reload_warm_sessions(monkeypatch):
+    """A drifted-high sessions.message_count must not reload on every read.
+
+    `_persist_message` swallows a failed insert while `add_message` has already
+    appended in memory, so the next successful persist writes rows+1. Keyed on
+    that column, the hydration gate would stay true forever and re-select the
+    whole transcript on every send, edit, delete and truncate.
+    """
+    engine, db_factory = _database()
+    _seed_session(db_factory, message_count=6, stored_count=8)
+
+    manager = _manager(db_factory, monkeypatch)
+    try:
+        session = manager.get_session("session-1")
+        cold_loads = manager.full_loads
+        for _ in range(3):
+            manager.get_session("session-1")
+    finally:
+        engine.dispose()
+
+    assert len(session.history) == 6
+    assert cold_loads == 1
+    assert manager.full_loads == 1
+
+
+def test_stale_low_message_count_column_still_hydrates_for_the_model(monkeypatch):
+    """The other drift direction must not hand the model a truncated transcript.
+
+    `_persist_message` writes message_count = 0 when the session is not cached.
+    A partly-filled cache plus that stale-low column previously left the send
+    path with whatever RAM happened to hold.
+    """
+    engine, db_factory = _database()
+    _seed_session(db_factory, message_count=6, stored_count=0)
+
+    manager = _manager(
+        db_factory,
+        monkeypatch,
+        sessions={
+            "session-1": Session(
+                id="session-1",
+                name="Long chat",
+                endpoint_url="http://model.test/v1",
+                model="test-model",
+                owner="alice",
+                history=[ChatMessage("user", "content-0")],
+                message_count=0,
+            )
+        },
+    )
+    try:
+        session = manager.get_session("session-1")
+        manager.get_session("session-1")
+    finally:
+        engine.dispose()
+
+    assert [message.content for message in session.history] == [
+        f"content-{index}" for index in range(6)
+    ]
+    assert manager.full_loads == 1
+
+
+def test_fork_after_restart_copies_the_real_transcript(monkeypatch):
+    """Forking reads source.history, so it must hydrate through get_session.
+
+    Display pagination no longer fills the cache, so a fork taken after a
+    restart used to return HTTP 200 with an empty conversation.
+    """
+    engine, db_factory = _database()
+    _seed_session(db_factory, message_count=6)
+
+    # Restart state: metadata-only cache entry, exactly what load_sessions seeds.
+    manager = _manager(db_factory, monkeypatch)
+    manager.load_sessions()
+
+    monkeypatch.setattr(history_routes, "SessionLocal", db_factory)
+    monkeypatch.setattr(history_routes, "_verify_session_owner", lambda *_args: None)
+    monkeypatch.setattr("core.models._SESSION_MANAGER_INSTANCE", manager)
+
+    app = FastAPI()
+    app.include_router(history_routes.setup_history_routes(manager))
+    client = TestClient(app)
+
+    try:
+        page = client.get("/api/history/session-1?limit=2")
+        assert page.status_code == 200
+        assert len(manager.sessions["session-1"].history) == 0
+
+        response = client.post("/api/session/session-1/fork", json={"keep_count": 4})
+        assert response.status_code == 200
+        payload = response.json()
+        assert payload["kept"] == 4
+
+        forked = manager.get_session(payload["id"])
+        assert [message.content for message in forked.history] == [
+            f"content-{index}" for index in range(4)
+        ]
+    finally:
+        engine.dispose()
+
+
+class _ContextBuildReached(Exception):
+    pass
+
+
+class _ToolPolicy:
+    block_all_tool_calls = False
+
+    def blocks(self, _tool_name):
+        return False
+
+
+class _ChatHandler:
+    async def handle_memory_command(self, _session, _message):
+        return None
+
+
+def _json_request(path, payload):
+    raw = json.dumps(payload).encode()
+    sent = False
+
+    async def receive():
+        nonlocal sent
+        if sent:
+            return {"type": "http.request", "body": b"", "more_body": False}
+        sent = True
+        return {"type": "http.request", "body": raw, "more_body": False}
+
+    scope = {
+        "type": "http",
+        "asgi": {"version": "3.0"},
+        "http_version": "1.1",
+        "method": "POST",
+        "scheme": "http",
+        "path": path,
+        "raw_path": path.encode(),
+        "root_path": "",
+        "query_string": b"",
+        "headers": [(b"content-type", b"application/json")],
+        "client": ("127.0.0.1", 1234),
+        "server": ("testserver", 80),
+    }
+    return Request(scope, receive)
+
+
+def _route_endpoint(router, path):
+    return next(route.endpoint for route in router.routes if route.path == path)
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("path", ["/api/chat", "/api/chat_stream"])
+async def test_model_send_routes_hydrate_before_context_build(monkeypatch, path):
+    # A real SessionManager over a real (temp) DB — a stub here would only
+    # assert that the stub hydrates, not that SessionManager does.
+    engine, db_factory = _database()
+    _seed_session(db_factory, message_count=6, stored_count=8)
+    manager = _manager(db_factory, monkeypatch)
+    manager.load_sessions()  # restart state: metadata only, no messages cached
+    contexts_built = []
+
+    async def assert_complete_context(session, *_args, **_kwargs):
+        contexts_built.append(session)
+        assert [message.content for message in session.history] == [
+            f"content-{index}" for index in range(6)
+        ]
+        raise _ContextBuildReached
+
+    monkeypatch.setattr(chat_routes, "_set_user_time_from_request", lambda *_args: None)
+    monkeypatch.setattr(chat_routes, "_verify_session_owner", lambda *_args: None)
+    monkeypatch.setattr(chat_routes, "effective_user", lambda *_args: "alice")
+    monkeypatch.setattr(
+        chat_routes,
+        "_clear_orphaned_session_endpoint",
+        lambda *_args, **_kwargs: False,
+    )
+    monkeypatch.setattr(
+        chat_routes,
+        "_recover_empty_session_model",
+        lambda *_args, **_kwargs: False,
+    )
+    monkeypatch.setattr(chat_routes, "_enforce_chat_privileges", lambda *_args: None)
+    monkeypatch.setattr(
+        chat_routes,
+        "build_effective_tool_policy",
+        lambda **_kwargs: _ToolPolicy(),
+    )
+    monkeypatch.setattr(chat_routes, "build_chat_context", assert_complete_context)
+    monkeypatch.setattr(
+        chat_routes,
+        "_resolve_request_workspace",
+        lambda *_args: (None, False),
+    )
+    monkeypatch.setattr(chat_routes, "_classify_tool_intent", lambda *_args: None)
+    monkeypatch.setattr(
+        chat_routes,
+        "_is_contextual_web_followup",
+        lambda *_args: False,
+    )
+    monkeypatch.setattr(
+        chat_routes,
+        "_is_contextual_browser_followup",
+        lambda *_args: False,
+    )
+    monkeypatch.setattr(
+        chat_routes,
+        "_resolve_workspace_from_message_path",
+        lambda *_args: (None, None),
+    )
+    monkeypatch.setattr(
+        chat_routes,
+        "_reconcile_selected_route_from_request",
+        lambda *_args, **_kwargs: False,
+    )
+    monkeypatch.setattr(chat_routes, "resolve_session_auth", lambda *_args, **_kwargs: None)
+    monkeypatch.setattr(chat_routes, "get_session_mode", lambda *_args: "chat")
+    monkeypatch.setattr(
+        chat_routes,
+        "_is_image_generation_session",
+        lambda *_args, **_kwargs: False,
+    )
+    monkeypatch.setattr(chat_routes, "web_search_enabled_for_turn", lambda *_args: False)
+
+    router = chat_routes.setup_chat_routes(
+        manager,
+        _ChatHandler(),
+        object(),
+        object(),
+        object(),
+        object(),
+    )
+    endpoint = _route_endpoint(router, path)
+
+    async def send():
+        if path == "/api/chat":
+            await endpoint(
+                _json_request(path, {}),
+                ChatRequest(message="hello", session="session-1"),
+            )
+        else:
+            await endpoint(
+                _json_request(
+                    path,
+                    {"message": "hello", "session": "session-1"},
+                )
+            )
+
+    try:
+        with pytest.raises(_ContextBuildReached):
+            await send()
+        first_loads = manager.full_loads
+
+        # Second send on the now-warm session: the transcript is complete, so
+        # it must be served from RAM even though sessions.message_count is
+        # still drifted high in the DB.
+        with pytest.raises(_ContextBuildReached):
+            await send()
+    finally:
+        engine.dispose()
+
+    assert len(contexts_built) == 2
+    assert contexts_built[0] is contexts_built[1]
+    assert first_loads == 1
+    assert manager.full_loads == 1