ROOT FIX: Override agent.py with correct OpenRouter models + routing

backend/routes/agent_header.py

@@ -0,0 +1,75 @@
+"""Minimal agent routes override — patches model list and LLM routing for OpenRouter."""
+
+# Import everything from the original module path (cloned source)
+import asyncio
+import json
+import logging
+import os
+import sys
+from typing import Any
+
+# Patch LLM params FIRST before any other import uses it
+from agent.core import llm_params as _lp
+_original_resolve = _lp._resolve_llm_params
+
+def _openrouter_resolve(model_name, session_hf_token=None, reasoning_effort=None, strict=False):
+    """Route all models through OpenRouter if configured."""
+    openrouter_key = os.environ.get("OPENROUTER_API_KEY") or os.environ.get("OPENAI_API_KEY")
+    openrouter_base = os.environ.get("OPENAI_API_BASE", "")
+    
+    if openrouter_key and "openrouter" in openrouter_base:
+        # Route through OpenRouter for ANY model
+        if model_name.startswith("openai/"):
+            raw = model_name[len("openai/"):]
+        elif model_name.startswith("anthropic/") or model_name.startswith("bedrock/"):
+            # Let anthropic/bedrock go through original if no openrouter configured
+            if not openrouter_key:
+                return _original_resolve(model_name, session_hf_token, reasoning_effort, strict)
+            raw = model_name
+        else:
+            raw = model_name
+        return {
+            "model": f"openai/{raw}",
+            "api_base": openrouter_base,
+            "api_key": openrouter_key,
+        }
+    
+    return _original_resolve(model_name, session_hf_token, reasoning_effort, strict)
+
+_lp._resolve_llm_params = _openrouter_resolve
+logging.getLogger(__name__).info("[patch] ✅ OpenRouter LLM routing active")
+
+# Now import the real routes module content
+from dependencies import INTERNAL_HF_TOKEN_KEY, get_current_user
+from fastapi import APIRouter, Depends, HTTPException, Request
+from fastapi.exceptions import RequestValidationError
+from fastapi.responses import StreamingResponse
+from huggingface_hub.errors import HfHubHTTPError
+from litellm import Message, acompletion
+from pydantic import ValidationError
+from starlette.datastructures import FormData, UploadFile
+from dataset_uploads import MAX_DATASET_UPLOAD_BYTES, dataset_context_note, push_dataset_upload_to_hub
+from models import (ApprovalRequest, DatasetUploadResponse, HealthResponse, LLMHealthResponse,
+                    SessionInfo, SessionNotificationsRequest, SessionResponse, SessionYoloRequest,
+                    SubmitRequest, TruncateRequest)
+from session_manager import MAX_SESSIONS, AgentSession, SessionCapacityError, session_manager
+import user_quotas
+from agent.core.hf_access import get_jobs_access
+from agent.core.hf_tokens import resolve_hf_request_token, resolve_hf_router_token
+from agent.core.llm_params import _resolve_llm_params
+
+logger = logging.getLogger(__name__)
+router = APIRouter(prefix="/api", tags=["agent"])
+_background_teardown_tasks: set[asyncio.Task] = set()
+
+# ====== PATCHED MODELS LIST ======
+DEFAULT_FREE_MODEL_ID = "deepseek/deepseek-chat-v3-0324"
+PREMIUM_MODEL_IDS = {"anthropic/claude-sonnet-4", "anthropic/claude-opus-4"}
+
+AVAILABLE_MODELS = [
+    {"id": "deepseek/deepseek-chat-v3-0324", "label": "DeepSeek V3", "provider": "openrouter", "tier": "free", "recommended": True},
+    {"id": "google/gemini-2.0-flash-001", "label": "Gemini 2.0 Flash", "provider": "openrouter", "tier": "free"},
+    {"id": "qwen/qwen3-30b-a3b", "label": "Qwen3 30B", "provider": "openrouter", "tier": "free"},
+    {"id": "anthropic/claude-sonnet-4", "label": "Claude Sonnet 4", "provider": "openrouter", "tier": "pro"},
+]
+# ====== END PATCHED ======