sync: 159 file da Baida98/AI@ac70b534 (2026-08-16 12:27 UTC) [deploy-all]

.env.example

@@ -18,7 +18,11 @@ RAILWAY_PROJECT_ID=YOUR_RAILWAY_PROJECT_ID_A
 SUPABASE_URL=
 SUPABASE_SERVICE_ROLE_KEY=
 GITHUB_TOKEN=
+# Hugging Face Router: endpoint OpenAI-compatible per inferenza.
 HF_TOKEN=
+HF_MODEL=Qwen/Qwen2.5-Coder-32B-Instruct
+# Pool opzionale: [{"profile":"primary","api_key":"...","model":"openai/gpt-oss-120b:fastest"}]
+HF_ROUTER_PROFILES_JSON=
 
 # ── 3. Quadrante B (HANDS - Collab/Failover) ─────────────────
 RAILWAY_TOKEN_B=

api/providers.py

@@ -1,5 +1,6 @@
 """backend/api/providers.py — Health, tools, status, AI health, heartbeat (S354)."""
 import os, asyncio, time, logging
+import requests
 from fastapi import APIRouter, Request
 from fastapi import Depends
 from .auth_guard import require_role, AuthRole
@@ -222,6 +223,44 @@ async def ai_provider_health(role: AuthRole = Depends(require_role(AuthRole.MACH
     from models.ai_client import AIClient
     client = AIClient()
 
+    def _classify_probe_error(exc: Exception) -> str:
+        message = str(exc).lower()
+        if "429" in message or "rate limit" in message or "quota" in message:
+            return "rate_limit_or_quota"
+        if "402" in message or "payment" in message or "credit" in message:
+            return "credits_exhausted"
+        if "401" in message or "403" in message or "unauthorized" in message or "forbidden" in message:
+            return "authentication_or_permission"
+        if "timeout" in message or "timed out" in message:
+            return "timeout"
+        if "404" in message or "not found" in message:
+            return "model_or_endpoint_not_found"
+        return "upstream_error"
+
+    async def _openrouter_key_limits(provider) -> dict:
+        if provider.name != "openrouter":
+            return {}
+        try:
+            response = await asyncio.to_thread(
+                requests.get,
+                "https://openrouter.ai/api/v1/key",
+                headers={"Authorization": f"Bearer {provider.api_key}"},
+                timeout=8,
+            )
+            body = response.json() if response.content else {}
+            data = body.get("data") if isinstance(body, dict) else {}
+            if response.status_code >= 400:
+                return {"key_status": response.status_code, "key_error_class": _classify_probe_error(RuntimeError(f"HTTP {response.status_code}"))}
+            return {
+                "key_status": response.status_code,
+                "limit_remaining": data.get("limit_remaining"),
+                "limit_reset": data.get("limit_reset"),
+                "is_free_tier": data.get("is_free_tier"),
+                "usage_daily": data.get("usage_daily"),
+            }
+        except Exception as exc:
+            return {"key_error_class": _classify_probe_error(exc)}
+
     async def _probe(provider) -> dict:
         t0 = time.monotonic()
         try:
@@ -237,12 +276,15 @@ async def ai_provider_health(role: AuthRole = Depends(require_role(AuthRole.MACH
                 timeout=8.0,
             )
             ms = round((time.monotonic() - t0) * 1000)
-            return {"name": provider.name, "ok": True, "status": "ok", "latency_ms": ms,
-                    "model": provider.default_model.split("/")[-1][:28]}
+            result = {"name": provider.name, "profile": provider.profile, "ok": True, "status": "ok", "latency_ms": ms,
+                      "model": provider.default_model.split("/")[-1][:40]}
+            result.update(await _openrouter_key_limits(provider))
+            return result
         except Exception as exc:
             ms = round((time.monotonic() - t0) * 1000)
-            return {"name": provider.name, "ok": False, "status": "error", "latency_ms": ms,
-                    "error": str(exc)[:300], "model": provider.default_model.split("/")[-1][:28]}  # S606: 200→300
+            return {"name": provider.name, "profile": provider.profile, "ok": False, "status": "error", "latency_ms": ms,
+                    "error_class": _classify_probe_error(exc),
+                    "error": str(exc)[:300], "model": provider.default_model.split("/")[-1][:40], **(await _openrouter_key_limits(provider))}
 
     results = list(await asyncio.gather(*[_probe(p) for p in client.providers]))
     payload = {"providers": results, "tested_at": int(time.time() * 1000)}

api/state.py

@@ -121,7 +121,7 @@ SENSITIVE = {
     'VITE_OPENROUTER_API_KEY', 'VITE_HF_TOKEN', 'VITE_GROQ_API_KEY',
     'OPENROUTER_PROFILES_JSON', 'GROQ_PROFILES_JSON', 'CEREBRAS_PROFILES_JSON',
     'SAMBANOVA_PROFILES_JSON', 'GEMINI_PROFILES_JSON', 'NVIDIA_PROFILES_JSON',
-    'HF_ROUTER_PROFILES_JSON',
+    'HF_ROUTER_PROFILES_JSON', 'HF_MODEL',
     'GH_PAGES_TOKEN', 'VERCEL_TOKEN',
 }
 

main.py

@@ -59,7 +59,7 @@ async def _run_auto_migration():
         'VITE_OPENROUTER_API_KEY', 'VITE_HF_TOKEN', 'VITE_GROQ_API_KEY',
         'OPENROUTER_PROFILES_JSON', 'GROQ_PROFILES_JSON', 'CEREBRAS_PROFILES_JSON',
         'SAMBANOVA_PROFILES_JSON', 'GEMINI_PROFILES_JSON', 'NVIDIA_PROFILES_JSON',
-        'HF_ROUTER_PROFILES_JSON',
+        'HF_ROUTER_PROFILES_JSON', 'HF_MODEL',
         'GH_PAGES_TOKEN', 'VERCEL_TOKEN'
     ]
     

models/ai_client.py

@@ -307,9 +307,9 @@ class AIClient:
         return self._execution_pool(candidates, f"fallback:{purpose}")
 
     async def _fetch_one(self, provider: ProviderConfig, messages: list, temperature: float, max_tokens: int) -> Tuple[ProviderConfig, str, float]:
-        client = self._client_for(provider)
         start = _time_mod.monotonic()
         try:
+            client = self._client_for(provider)
             response = await asyncio.wait_for(
                 asyncio.to_thread(
                     client.chat.completions.create, 
@@ -458,9 +458,9 @@ class AIClient:
         attempted: list[str] = []
         for provider in providers:
             attempted.append(provider.name)
-            client = self._client_for(provider)
             emitted = False
             try:
+                client = self._client_for(provider)
                 stream = await asyncio.to_thread(
                     client.chat.completions.create,
                     model=provider.default_model,