Update README.md

README.md

@@ -11,33 +11,15 @@ license: mit
 
 # Yappy — AI backend
 
-Yappy is an AI companion that motivates, builds discipline, supports, or
-just chats — depending on the mode you pick. Created by Rik.
+Uses Hugging Face's hosted Inference API instead of running a model
+locally — avoids overloading the free CPU Space.
 
-This Space runs the backend: a FastAPI server wrapping Qwen2.5-3B-Instruct
-(GGUF, via `llama-cpp-python`) on CPU, with streaming responses. The
-frontend (PWA) is hosted separately on Vercel and calls this Space's API.
+## Required secrets
 
-## Memory model
-
-**This server is stateless.** It holds no conversation history of its
-own. The frontend (browser localStorage) owns the chat history and sends
-the relevant messages back with every request in the `history` field.
-
-- Chat history survives the Space sleeping or restarting.
-- Each mode (motivation/discipline/support/chat) should keep its own
-  separate history array client-side and send only that mode's history.
-- The server does no per-user storage, no database, no cookies.
-
-## Modes
-
-| Mode | Behavior |
-|---|---|
-| `motivation` | Roasts you hard to get you moving |
-| `discipline` | Serious, lays out consequences of success vs failure |
-| `support` | Soft, validating, listens first |
-| `chat` | Just a casual friendly conversation |
+- **HF_TOKEN** (required) — HF access token with Inference API permissions
+- **YAPPY_API_KEY** (optional) — shared secret for request auth
 
 ## API
 
-All endpoints (except `/` and `/health`) expect header:
+POST /chat/stream — SSE stream, events: thinking, token, done
+POST /chat — non-streaming, returns { reply, thinking }