Update main.py

main.py

@@ -61,8 +61,8 @@ DEFAULT_AGENTS = [
          "5. Si hay frontend_dev en el equipo, TÚ haces servidor/backend, él hace HTML.\n"
          "6. Si la tarea no requiere backend → responde: {\"skip\":\"no backend needed\"}"
      ),
-     "models":["qwen/qwen3-4b:free","meta-llama/llama-3.3-70b-instruct:free",
-               "mistralai/mistral-small-3.1-24b-instruct:free","google/gemma-3-12b-it:free"]},
+     "models":["google/gemma-3-27b-it:free","google/gemma-3-12b-it:free",
+               "meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free"]},
     {"key":"frontend_dev","name":"Frontend","provider":"openrouter",
      "role":(
          "Eres desarrollador frontend senior. REGLAS ABSOLUTAS:\n"
@@ -71,8 +71,8 @@ DEFAULT_AGENTS = [
          "3. Si la tarea NO requiere frontend → responde: {\"skip\":\"no frontend needed\"}\n"
          "4. Entrega siempre HTML completo y funcional con los estilos incluidos."
      ),
-     "models":["meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free",
-               "qwen/qwen3-4b:free","google/gemma-3-12b-it:free"]},
+     "models":["google/gemma-3-12b-it:free","google/gemma-3-27b-it:free",
+               "meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free"]},
     {"key":"analyst","name":"Analyst","provider":"openrouter",
      "role":(
          "Eres analista de negocios. REGLAS:\n"
@@ -80,8 +80,8 @@ DEFAULT_AGENTS = [
          "2. NUNCA describas imágenes ni hagas trabajo de otros agentes.\n"
          "3. Si la tarea no requiere análisis → responde: {\"skip\":\"no analysis needed\"}"
      ),
-     "models":["meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free",
-               "google/gemma-3-27b-it:free","qwen/qwen3-4b:free"]},
+     "models":["google/gemma-3-27b-it:free","google/gemma-3-12b-it:free",
+               "meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free"]},
     {"key":"writer","name":"Writer","provider":"openrouter",
      "role":(
          "Eres redactor experto. Escribe SOLO contenido real y extenso (500+ palabras). "
@@ -89,8 +89,8 @@ DEFAULT_AGENTS = [
          "Secciones: ## Resumen Ejecutivo, ### Introducción, ### Desarrollo, "
          "### Hallazgos, ### Conclusiones, ### Recomendaciones"
      ),
-     "models":["meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free",
-               "qwen/qwen3-4b:free","google/gemma-3-12b-it:free"]},
+     "models":["google/gemma-3-12b-it:free","google/gemma-3-27b-it:free",
+               "meta-llama/llama-3.3-70b-instruct:free","mistralai/mistral-small-3.1-24b-instruct:free"]},
     {"key":"image_agent","name":"ImageAgent","provider":"gemini",
      "role":(
          "Cuando se te pida imágenes, responde SOLO con: "
@@ -117,6 +117,87 @@ async def call_compat(base_url,model,system,user,key,headers):
         r.raise_for_status()
         return r.json()["choices"][0]["message"]["content"]
 
+
+async def call_compat_multiturn(base_url, model, system, messages, key, extra_headers):
+    """OpenAI-compatible chat with full message history for multi-turn conversations."""
+    h = {"Authorization": f"Bearer {key}", "Content-Type": "application/json", **extra_headers}
+    payload = {
+        "model": model,
+        "messages": [{"role": "system", "content": system}] + messages,
+        "max_tokens": 2048,
+        "temperature": 0.6,
+    }
+    async with httpx.AsyncClient(timeout=90) as c:
+        r = await c.post(base_url, json=payload, headers=h)
+        r.raise_for_status()
+        return r.json()["choices"][0]["message"]["content"]
+
+async def call_gemini_multiturn(model, system, messages, key):
+    """Gemini multi-turn conversation."""
+    url = f"https://generativelanguage.googleapis.com/v1beta/models/{model}:generateContent?key={key}"
+    # Convert messages to Gemini format
+    contents = []
+    for m in messages:
+        role = "user" if m["role"] == "user" else "model"
+        contents.append({"role": role, "parts": [{"text": m["content"]}]})
+    # Prepend system as first user message if contents start with model
+    full_system = system + "\n\n" + (contents[0]["parts"][0]["text"] if contents and contents[0]["role"] == "user" else "")
+    if contents and contents[0]["role"] == "user":
+        contents[0]["parts"][0]["text"] = full_system
+    payload = {
+        "contents": contents,
+        "generationConfig": {"maxOutputTokens": 2048, "temperature": 0.6},
+    }
+    async with httpx.AsyncClient(timeout=90) as c:
+        r = await c.post(url, json=payload)
+        r.raise_for_status()
+        return r.json()["candidates"][0]["content"]["parts"][0]["text"]
+
+async def call_llm_multiturn(agent, messages):
+    """Multi-turn LLM call with full conversation history. Cascades through providers."""
+    system = agent["role"]
+    last_err = None
+
+    # 1. Primary provider
+    p = PROVIDERS[agent["provider"]]
+    for m in agent["models"]:
+        try:
+            if p["type"] == "gemini":
+                return await call_gemini_multiturn(m, system, messages, p["key"])
+            else:
+                return await call_compat_multiturn(p["base_url"], m, system, messages,
+                                                    p["key"], p.get("headers", {}))
+        except Exception as e:
+            last_err = str(e)
+            if is_rate_limit(last_err):
+                break
+
+    # 2. OpenRouter fallback (Gemma 3 first)
+    if OPENROUTER_API_KEY and agent["provider"] != "openrouter":
+        or_prov = PROVIDERS["openrouter"]
+        for m in ["google/gemma-3-27b-it:free", "google/gemma-3-12b-it:free",
+                  "meta-llama/llama-3.3-70b-instruct:free",
+                  "mistralai/mistral-small-3.1-24b-instruct:free"]:
+            try:
+                return await call_compat_multiturn(or_prov["base_url"], m, system, messages,
+                                                    or_prov["key"], or_prov.get("headers", {}))
+            except Exception as e:
+                last_err = str(e)
+                if is_rate_limit(last_err):
+                    break
+
+    # 3. Groq fallback
+    if GROQ_API_KEY and agent["provider"] != "groq":
+        groq = PROVIDERS["groq"]
+        for m in ["llama-3.1-8b-instant", "gemma2-9b-it"]:
+            try:
+                return await call_compat_multiturn(groq["base_url"], m, system, messages,
+                                                    GROQ_API_KEY, {})
+            except Exception as e:
+                last_err = str(e)
+
+    raise Exception(f"All providers exhausted. Last: {last_err}")
+
 def is_rate_limit(err: str) -> bool:
     e = err.lower()
     return any(x in e for x in ["429","rate limit","quota","resource exhausted","too many requests","ratelimit"])
@@ -144,12 +225,12 @@ async def call_llm(agent, task):
     if OPENROUTER_API_KEY and agent["provider"] != "openrouter":
         or_prov = PROVIDERS["openrouter"]
         for m in [
+                  "google/gemma-3-27b-it:free",
+                  "google/gemma-3-12b-it:free",
                   "meta-llama/llama-3.3-70b-instruct:free",
                   "mistralai/mistral-small-3.1-24b-instruct:free",
                   "qwen/qwen3-4b:free",
-                  "google/gemma-3-12b-it:free",
                   "qwen/qwen-2.5-72b-instruct:free",
-                  "microsoft/phi-4-reasoning-plus:free",
                   "deepseek/deepseek-r1-distill-llama-70b:free",
                   ]:
             try:
@@ -518,25 +599,66 @@ async def run_mission(request:Request):
 
 
 
+# In-memory chat sessions: {session_id: [{role, content}]}
+chat_sessions: dict = {}
+
 @app.post("/api/chat")
 async def chat_with_agent(request: Request):
-    body = await request.json()
-    agent_key = body.get("agent", "").strip()
-    message   = body.get("message", "").strip()
+    body        = await request.json()
+    agent_key   = body.get("agent", "").strip()
+    message     = body.get("message", "").strip()
+    session_id  = body.get("session_id", agent_key)   # default: one session per agent
+    clear       = body.get("clear", False)
+
     if not agent_key or not message:
         return JSONResponse({"error": "agent and message required"}, status_code=400)
     if agent_key not in agent_registry:
         return JSONResponse({"error": f"Agent '{agent_key}' not found"}, status_code=404)
-    from datetime import datetime as _dt
-    _today = _dt.now().strftime("%A %d de %B de %Y, %H:%M")
+
+    # Reset history if requested
+    if clear:
+        chat_sessions[session_id] = []
+
+    # Init session
+    if session_id not in chat_sessions:
+        chat_sessions[session_id] = []
+
+    # Build messages list (keep last 20 turns = 40 messages to stay within context)
+    history = chat_sessions[session_id][-40:]
+    history.append({"role": "user", "content": message})
+
+    # Inject today's date into agent role
+    _today = datetime.now().strftime("%A %d de %B de %Y, %H:%M")
     agent = dict(agent_registry[agent_key])
-    agent["role"] = f"HOY ES: {_today}. " + agent["role"]
+    agent["role"] = f"HOY ES: {_today}. Eres {agent['name']}. " + agent["role"]
+
     try:
-        response = await call_llm(agent, message)
-        return JSONResponse({"success": True, "agent": agent_key, "response": response, "model": agent["models"][0] if agent.get("models") else ""})
+        response = await call_llm_multiturn(agent, history)
+
+        # Save turn to session
+        chat_sessions[session_id].append({"role": "user",      "content": message})
+        chat_sessions[session_id].append({"role": "assistant",  "content": response})
+
+        # Keep sessions from growing too large (max 100 messages)
+        if len(chat_sessions[session_id]) > 100:
+            chat_sessions[session_id] = chat_sessions[session_id][-80:]
+
+        used_model = agent["models"][0] if agent.get("models") else ""
+        return JSONResponse({
+            "success":    True,
+            "agent":      agent_key,
+            "response":   response,
+            "model":      used_model,
+            "turn":       len(chat_sessions[session_id]) // 2,
+        })
     except Exception as e:
         return JSONResponse({"error": str(e)}, status_code=500)
 
+@app.delete("/api/chat/{session_id}")
+async def clear_chat_session(session_id: str):
+    chat_sessions.pop(session_id, None)
+    return {"success": True}
+
 @app.get("/api/archive")
 async def list_archive():
     files = []