sync: 141 file da Baida98/AI@c6afe8eb (2026-08-07 20:01 UTC) [deploy-all]

api/state.py

@@ -137,6 +137,23 @@ _heartbeat_state: dict = {
     "runs":            0,
 }
 
+# ── Telemetry & Timing ────────────────────────────────────────────────────────
+# Shared by the agent loop and the provider diagnostics endpoint. Keep this
+# bounded so long-running workers cannot grow without limit.
+_TIMING_STORE: dict[str, list[float]] = {}
+_REPAIR_STATS: dict[str, int] = {}
+
+def record_timing(key: str, duration_ms: float) -> None:
+    """Record a bounded latency sample for agent/provider diagnostics."""
+    samples = _TIMING_STORE.setdefault(key, [])
+    samples.append(duration_ms)
+    if len(samples) > 100:
+        samples.pop(0)
+
+def increment_stat(key: str, delta: int = 1) -> None:
+    """Increment an aggregated agent quality/recovery counter."""
+    _REPAIR_STATS[key] = _REPAIR_STATS.get(key, 0) + delta
+
 # ── Singleton Getters ─────────────────────────────────────────────────────────
 def get_supabase() -> Optional[Any]:
     return _sb
@@ -147,10 +164,16 @@ def _get_mem_manager() -> Any:
     global _mem_manager, _mem_manager_inited
     if _mem_manager_inited: return _mem_manager
     try:
-        from memory.memory_manager import MemoryManager
-        _mem_manager = MemoryManager()
-        _mem_manager_inited = True
-    except Exception: _mem_manager = None
+        from memory.manager import MemoryManager
+        _mem_manager = MemoryManager(sb_client=_get_sb())
+        try:
+            _asyncio_mod.create_task(_mem_manager.init())
+            _mem_manager_inited = True
+        except RuntimeError:
+            # No running event loop during import; the async getter initializes it.
+            pass
+    except Exception:
+        _mem_manager = None
     return _mem_manager
 
 _executor: Any = None
@@ -175,3 +198,97 @@ def _get_ai_client() -> Any:
 
 async def _get_mem_manager_async() -> Any:
     return _get_mem_manager()
+
+_planner: Any = None
+def _get_planner() -> Any:
+    global _planner
+    if _planner is not None: return _planner
+    try:
+        from agents.planner import Planner
+        _planner = Planner(llm_client=_get_ai_client())
+    except Exception:
+        _planner = None
+    return _planner
+
+# ── Prune helpers ─────────────────────────────────────────────────────────────
+def _prune_checkpoints() -> None:
+    now = int(time.time() * 1000)
+    expired = [k for k, v in list(_task_checkpoints.items())
+               if now - v.get('savedAt', 0) > _CHECKPOINT_TTL_MS]
+    for k in expired:
+        _task_checkpoints.pop(k, None)
+    if len(_task_checkpoints) > _CHECKPOINT_MAX:
+        oldest = sorted(_task_checkpoints.items(), key=lambda x: x[1].get('savedAt', 0))
+        for k, _ in oldest[:len(_task_checkpoints) - _CHECKPOINT_MAX]:
+            _task_checkpoints.pop(k, None)
+
+def _prune_agent_tasks() -> None:
+    now = int(time.time() * 1000)
+    expired = [k for k, v in list(_agent_tasks.items())
+               if v.get('status') in ('SUCCESS', 'ERROR', 'CANCELLED')
+               and now - v.get('created_at', 0) > _AGENT_TASK_TTL_MS]
+    for k in expired:
+        _agent_tasks.pop(k, None)
+    if len(_agent_tasks) > _AGENT_TASK_MAX:
+        oldest = sorted(_agent_tasks.items(), key=lambda x: x[1].get('created_at', 0))
+        for k, _ in oldest[:len(_agent_tasks) - _AGENT_TASK_MAX]:
+            _agent_tasks.pop(k, None)
+
+def _prune_loop_registry() -> None:
+    now = time.time()
+    stale = [k for k, v in list(_loop_registry.items())
+             if v.get('done') and now - v.get('finished_at', 0.0) > _LOOP_REGISTRY_TTL_S]
+    for k in stale:
+        _loop_registry.pop(k, None)
+
+# ── Shared Pydantic models ────────────────────────────────────────────────────
+class ReasonLoopIn(BaseModel):
+    goal: str
+    context: list[dict] = []
+    max_steps: int = 8
+    project_context: str = ""
+    learning_hints: list[str] = []
+    session_id: Optional[str] = None
+    negative_constraints: Optional[str] = ""
+
+    @field_validator('goal', mode='before')
+    @classmethod
+    def validate_goal(cls, v: object) -> str:
+        if not isinstance(v, str) or not v.strip():
+            raise ValueError('goal must be a non-empty string')
+        return v.strip()
+
+    @field_validator('context', 'learning_hints', mode='before')
+    @classmethod
+    def coerce_list(cls, v: object) -> list:
+        return v if isinstance(v, list) else []
+
+    @field_validator('project_context', mode='before')
+    @classmethod
+    def coerce_str(cls, v: object) -> str:
+        return str(v).strip()[:2000] if v else ""
+
+class AgentTaskIn(BaseModel):
+    goal: str
+    context: list[dict] = []
+    max_steps: int = 8
+    taskId: Optional[str] = None
+    project_context: str = ""
+    learning_hints: list[str] = []
+    session_id: Optional[str] = None
+    resume_from_step: Optional[int] = None
+    persona: Optional[str] = None
+    negative_constraints: Optional[str] = ""
+
+    @field_validator('goal', mode='before')
+    @classmethod
+    def validate_goal(cls, v: object) -> str:
+        if not isinstance(v, str) or not v.strip():
+            raise ValueError('goal must be a non-empty string')
+        return v.strip()
+
+    @field_validator('context', 'learning_hints', mode='before')
+    @classmethod
+    def coerce_list(cls, v: object) -> list:
+        return v if isinstance(v, list) else []
+