Merge remote-tracking branch 'origin/main' into codex/im-channel-connections

# Conflicts:
#	backend/app/gateway/services.py
#	frontend/src/app/workspace/chats/page.tsx

backend/app/gateway/app.py

@@ -185,21 +185,27 @@ async def lifespan(app: FastAPI) -> AsyncGenerator[None, None]:
     # Pre-warm tiktoken encoding cache so the first memory-injection request
     # never blocks on the BPE data download (which hits an OpenAI/Azure URL
     # that may be unreachable in restricted networks — see issue #3402).
-    try:
-        from deerflow.agents.memory.prompt import warm_tiktoken_cache
+    # When memory.token_counting is "char", token counting never touches
+    # tiktoken, so skip the warm-up entirely (avoids even the 5s probe in
+    # network-restricted deployments — see issue #3429).
+    if startup_config.memory.token_counting == "char":
+        logger.info("memory.token_counting='char'; skipping tiktoken warm-up (network-free token estimation)")
+    else:
+        try:
+            from deerflow.agents.memory.prompt import warm_tiktoken_cache
 
-        warmed = await asyncio.wait_for(
-            asyncio.to_thread(warm_tiktoken_cache),
-            timeout=5,
-        )
-        if warmed:
-            logger.info("tiktoken encoding cache warmed successfully")
-        else:
-            logger.warning("tiktoken encoding cache warm-up failed; token counting will use character-based fallback")
-    except TimeoutError:
-        logger.warning("tiktoken encoding cache warm-up timed out; token counting will use character-based fallback")
-    except Exception:
-        logger.warning("tiktoken warm-up skipped", exc_info=True)
+            warmed = await asyncio.wait_for(
+                asyncio.to_thread(warm_tiktoken_cache),
+                timeout=5,
+            )
+            if warmed:
+                logger.info("tiktoken encoding cache warmed successfully")
+            else:
+                logger.warning("tiktoken encoding cache warm-up failed; token counting will use character-based fallback until tiktoken loads successfully")
+        except TimeoutError:
+            logger.warning("tiktoken encoding cache warm-up timed out; token counting will use character-based fallback until tiktoken loads successfully")
+        except Exception:
+            logger.warning("tiktoken warm-up skipped", exc_info=True)
 
     # Initialize LangGraph runtime components (StreamBridge, RunManager, checkpointer, store)
     async with langgraph_runtime(app, startup_config):

backend/app/gateway/routers/memory.py

@@ -98,6 +98,7 @@ class MemoryConfigResponse(BaseModel):
     fact_confidence_threshold: float = Field(..., description="Minimum confidence threshold for facts")
     injection_enabled: bool = Field(..., description="Whether memory injection is enabled")
     max_injection_tokens: int = Field(..., description="Maximum tokens for memory injection")
+    token_counting: str = Field(..., description="Token counting strategy for memory injection ('tiktoken' or 'char')")
 
 
 class MemoryStatusResponse(BaseModel):
@@ -310,7 +311,8 @@ async def get_memory_config_endpoint() -> MemoryConfigResponse:
             "max_facts": 100,
             "fact_confidence_threshold": 0.7,
             "injection_enabled": true,
-            "max_injection_tokens": 2000
+            "max_injection_tokens": 2000,
+            "token_counting": "tiktoken"
         }
         ```
     """
@@ -323,6 +325,7 @@ async def get_memory_config_endpoint() -> MemoryConfigResponse:
         fact_confidence_threshold=config.fact_confidence_threshold,
         injection_enabled=config.injection_enabled,
         max_injection_tokens=config.max_injection_tokens,
+        token_counting=config.token_counting,
     )
 
 
@@ -351,6 +354,7 @@ async def get_memory_status() -> MemoryStatusResponse:
             fact_confidence_threshold=config.fact_confidence_threshold,
             injection_enabled=config.injection_enabled,
             max_injection_tokens=config.max_injection_tokens,
+            token_counting=config.token_counting,
         ),
         data=MemoryResponse(**memory_data),
     )

backend/app/gateway/services.py

@@ -318,6 +318,21 @@ async def start_run(
             )
 
     owner_user_id = get_trusted_internal_owner_user_id(request)
+    # Stateless run endpoints carry thread_id in the request *body*, so the
+    # @require_permission(owner_check=True) decorator -- which resolves ownership
+    # from the path param -- cannot protect them. Enforce thread ownership here,
+    # before any run is created, so one user cannot start runs on (or read /wait
+    # checkpoint state from) another user's thread. Missing rows (auto-created
+    # temp threads) and NULL-owner rows (shared / pre-auth data) stay accessible
+    # via check_access; only a thread already owned by another user is rejected
+    # with 404, matching thread_runs.py's anti-enumeration behaviour. Internal
+    # channel runs act on behalf of IM users they do not own (see
+    # inject_authenticated_user_context), so the internal system role is exempt.
+    user = getattr(request.state, "user", None)
+    if user is not None and getattr(user, "system_role", None) != INTERNAL_SYSTEM_ROLE:
+        if not await run_ctx.thread_store.check_access(thread_id, str(user.id)):
+            raise HTTPException(status_code=404, detail=f"Thread {thread_id} not found")
+
     owner_context_token = set_current_user(SimpleNamespace(id=owner_user_id)) if owner_user_id else None
     try:
         try: