Update app.py

app.py

@@ -33,7 +33,7 @@ ALL_PROVIDERS = [
     {
         "name": "gemini",
         "base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
-        "model": "gemini-2.0-flash",
+        "model": "gemini-2.5-flash",
         "daily_limit": 1500,
         "priority": 2,
     },
@@ -68,17 +68,28 @@ ACTIVE_PROVIDERS = [
 
 # ── Usage tracker ──
 usage_count = defaultdict(int)
+bad_keys = set()       # permanently skip these for the session (bad key)
 last_reset = time.time()
 
-def pick_provider():
+def pick_provider(exclude=None):
     global last_reset
-    if time.time() - last_reset > 86400:  # reset every 24h
+    if exclude is None:
+        exclude = set()
+
+    if time.time() - last_reset > 86400:  # reset counts every 24h
         usage_count.clear()
+        bad_keys.clear()
         last_reset = time.time()
 
-    # Pick provider with most remaining capacity, respecting priority
+    available = [
+        p for p in ACTIVE_PROVIDERS
+        if p["name"] not in exclude and p["name"] not in bad_keys
+    ]
+    if not available:
+        return None
+
     return min(
-        ACTIVE_PROVIDERS,
+        available,
         key=lambda p: (
             usage_count[p["name"]] / p["daily_limit"],  # % used
             p["priority"]                                # tiebreak by priority
@@ -92,14 +103,7 @@ def respond(message, history):
         yield "❌ No API keys provided! Add them in your Space's Settings → Variables and Secrets."
         return
 
-    provider = pick_provider()
-    print(f"Using: {provider['name']} | uses today: {usage_count[provider['name']]}")
-
-    client = OpenAI(
-        api_key=API_KEYS[provider["name"]],
-        base_url=provider["base_url"],
-    )
-
+    # Build messages once — reused across retries
     messages = [{"role": "system", "content": SYSTEM_PROMPT}]
     for exchange in history[-3:]:
         user_msg = exchange[0] if isinstance(exchange, (list, tuple)) else ""
@@ -110,32 +114,50 @@ def respond(message, history):
             messages.append({"role": "assistant", "content": asst_msg})
     messages.append({"role": "user", "content": message})
 
-    try:
-        stream = client.chat.completions.create(
-            model=provider["model"],
-            messages=messages,
-            max_tokens=200,
-            temperature=0.7,
-            stream=True,
+    tried = set()  # skip these for this request only
+
+    while True:
+        provider = pick_provider(exclude=tried)
+
+        if provider is None:
+            yield "❌ All providers failed or hit limits. Try again later."
+            return
+
+        print(f"Trying: {provider['name']} | uses today: {usage_count[provider['name']]}")
+
+        client = OpenAI(
+            api_key=API_KEYS[provider["name"]],
+            base_url=provider["base_url"],
         )
-        usage_count[provider["name"]] += 1
-
-        response_text = ""
-        for chunk in stream:
-            token = chunk.choices[0].delta.content or ""
-            response_text += token
-            yield response_text
-
-    except Exception as e:
-        err = str(e).lower()
-        if "401" in err or "unauthorized" in err or "invalid api key" in err:
-            yield f"❌ API key for **{provider['name']}** is invalid or expired!"
-        elif "429" in err or "rate limit" in err:
-            yield f"⚠️ **{provider['name']}** hit rate limit — try again in a moment"
-        elif "quota" in err or "exceeded" in err:
-            yield f"⚠️ **{provider['name']}** daily quota exceeded"
-        else:
-            yield f"⚠️ **{provider['name']}** error: {str(e)}"
+
+        try:
+            stream = client.chat.completions.create(
+                model=provider["model"],
+                messages=messages,
+                max_tokens=200,
+                temperature=0.7,
+                stream=True,
+            )
+            usage_count[provider["name"]] += 1
+
+            response_text = ""
+            for chunk in stream:
+                token = chunk.choices[0].delta.content or ""
+                response_text += token
+                yield response_text
+            return  # success — stop retrying
+
+        except Exception as e:
+            err = str(e).lower()
+            tried.add(provider["name"])
+
+            if "401" in err or "unauthorized" in err or "invalid api key" in err:
+                bad_keys.add(provider["name"])  # skip for whole session
+                print(f"❌ {provider['name']} bad key — skipping for session")
+            elif "429" in err or "rate limit" in err or "quota" in err or "exceeded" in err:
+                print(f"⚠️ {provider['name']} rate limited — trying next provider")
+            else:
+                print(f"⚠️ {provider['name']} error: {e} — trying next provider")
 
 # ── Startup summary ──
 print("\n=== Provider Status ===")
@@ -162,4 +184,4 @@ demo = gr.ChatInterface(
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()