dynamic model selection..

app.py

@@ -313,6 +313,10 @@ def chat(request: ChatRequest, req: Request):
                     "Free credits exhausted. Please add credits at openrouter.ai/settings/credits "
                     "or update your API key in Settings."
                 )
+            elif "404" in str(e) or "not found" in error_str or "no endpoints" in error_str:
+                yield sse_error(
+                    "The AI model is temporarily unavailable. Please try again in a moment."
+                )
             elif "401" in str(e) or "unauthorized" in error_str or "invalid" in error_str:
                 yield sse_error(
                     "API key issue. Please check your OpenRouter API key in Settings."

graph.py

@@ -1,9 +1,7 @@
 """
-LangGraph definition — single chat node with free OpenRouter model selection.
-Checkpointer: SQLite via official LangGraph SqliteSaver.
-
-Text model:  nvidia/llama-3.1-nemotron-ultra-253b-v1:free
-Vision model: nex-agi/nex-n2-pro:free
+LangGraph definition — single chat node with dynamic free OpenRouter model selection.
+Fetches available free models from OpenRouter API, caches for 10 minutes.
+Defaults to openai/gpt-oss-120b:free for text, nex-agi/nex-n2-pro:free for vision.
 """
 
 from langgraph.graph import StateGraph, START, END
@@ -14,6 +12,10 @@ from langchain_openrouter import ChatOpenRouter
 from typing import TypedDict, Annotated
 
 import sqlite3
+import time
+import json
+import urllib.request
+import threading
 from langgraph.checkpoint.sqlite import SqliteSaver
 from config import OPENROUTER_API_KEY, DB_PATH
 import prompts
@@ -24,24 +26,117 @@ _conn = sqlite3.connect(DB_PATH, check_same_thread=False)
 checkpointer = SqliteSaver(conn=_conn)
 
 
-# --- Free model config ---
-TEXT_MODEL   = "nvidia/llama-3.1-nemotron-ultra-253b-v1:free"
-VISION_MODEL = "nex-agi/nex-n2-pro:free"
+# --- Dynamic free model selection ---
+
+DEFAULT_TEXT_MODEL  = "openai/gpt-oss-120b:free"
+DEFAULT_VISION_MODEL = "nex-agi/nex-n2-pro:free"
+
+_models_cache = None
+_models_cache_at = 0
+_models_lock = threading.Lock()
+MODELS_TTL = 10 * 60  # 10 minutes
+
+
+def _fetch_free_models() -> list[dict]:
+    """Fetch available free models from OpenRouter API."""
+    global _models_cache, _models_cache_at
+
+    with _models_lock:
+        if _models_cache and (time.time() - _models_cache_at) < MODELS_TTL:
+            return _models_cache
+
+    try:
+        req = urllib.request.Request(
+            "https://openrouter.ai/api/v1/models",
+            headers={"HTTP-Referer": "https://stemcopilot.app", "User-Agent": "STEMCopilot/1.0"},
+        )
+        with urllib.request.urlopen(req, timeout=8) as resp:
+            data = json.loads(resp.read().decode())
+
+        all_models = [
+            {"id": m["id"], "name": m.get("name", m["id"])}
+            for m in data.get("data", [])
+            if m.get("id", "").endswith(":free")
+        ]
+
+        # Separate text and vision-capable models
+        with _models_lock:
+            _models_cache = all_models
+            _models_cache_at = time.time()
+
+        return all_models
+
+    except Exception as e:
+        print(f"[MODELS] Could not fetch free models: {e}")
+        return _models_cache or []
+
+
+def _pick_text_model() -> str:
+    """Pick the best free text model. Prefers GPT OSS 120B, then NVIDIA nemotron models."""
+    models = _fetch_free_models()
+    model_ids = {m["id"] for m in models}
+
+    # Priority order
+    preferred = [
+        DEFAULT_TEXT_MODEL,
+        "openai/gpt-oss-120b:free",
+    ]
+
+    # Check preferred first
+    for mid in preferred:
+        if mid in model_ids:
+            return mid
+
+    # Then any NVIDIA model
+    nvidia = [m["id"] for m in models if m["id"].startswith("nvidia/")]
+    if nvidia:
+        return nvidia[0]
+
+    # Then any free model that isn't tiny
+    if models:
+        return models[0]["id"]
+
+    # Ultimate fallback
+    return DEFAULT_TEXT_MODEL
+
+
+def _pick_vision_model() -> str:
+    """Pick the best free vision/multimodal model."""
+    models = _fetch_free_models()
+    model_ids = {m["id"] for m in models}
+
+    # Priority order for vision
+    preferred = [
+        DEFAULT_VISION_MODEL,
+        "nex-agi/nex-n2-pro:free",
+    ]
+
+    for mid in preferred:
+        if mid in model_ids:
+            return mid
+
+    # Fallback to text model — it may handle images poorly but won't 404
+    return _pick_text_model()
+
+
+# Expose for app.py logging
+TEXT_MODEL = DEFAULT_TEXT_MODEL
+VISION_MODEL = DEFAULT_VISION_MODEL
 
 
 def _get_llm(api_key: str = "", model: str = "", has_image: bool = False):
-    """Create an LLM instance. Picks vision model if image is attached."""
+    """Create an LLM instance. Dynamically picks the best available free model."""
     key = api_key or OPENROUTER_API_KEY
     if not key:
         raise ValueError("No API key available. Please add your OpenRouter key in Settings.")
 
-    # Model priority: explicit override > auto-select based on image
+    # Model priority: explicit override > dynamic selection
     if model:
         mdl = model
     elif has_image:
-        mdl = VISION_MODEL
+        mdl = _pick_vision_model()
     else:
-        mdl = TEXT_MODEL
+        mdl = _pick_text_model()
 
     return ChatOpenRouter(
         model=mdl,