{"name":"Evolva Unified Gateway","version":"3.2.0","status":"running","modules":["health","auth","agents","memory","tools","inference","billing","citation"],"mcp_tools":7,"docs":"/docs","health":"/health","mcp":"/mcp/tools","auth_v2":"/auth/key","billing":"/billing/status","changelog":{"3.1.0":["Open-weights model routing — Gemma 3 (1B/4B/12B/27B), Nemotron-51B, Llama 3.3 70B, Qwen2.5 72B, Phi-4, Mistral 7B","HF Inference API backend — serverless open-weights calls via HF_TOKEN","Task-type-aware routing: /inference/route selects optimal model by task (understand/plan/code/chat/fast/long)","/inference/models endpoint lists all routes with open_weights flag and availability","Graceful fallback: if HF Inference API fails, routes to OpenAI-compatible backend"],"3.0.0":["Memory upgraded to DB-backed persistence (SQLite on /data or TiDB/MySQL via DATABASE_URL)","gateway_memory table: key, value, session_id, created_at, updated_at, ttl, expires_at, tags","gateway_api_keys, gateway_usage, gateway_billing tables — all data survives restarts","Auto-detects backend: SQLite (default) or MySQL/TiDB (when DATABASE_URL is set)","WAL mode enabled on SQLite for concurrent read/write safety","Startup event initialises schema on first boot"],"2.0.1":["Fix model routing — use gpt-4.1-mini as default"],"2.0.0":["Inference wired to real OpenAI-compatible LLM backend","Auth v2: /auth/key replaces session tokens","Billing module added"]}}