Atlas AI
Lightweight LLM running on local GPU accelerated hardware. Anyone can chat without auth. The client streams responses, shows round-trip latency for each turn, and we're training an Atlas-aware model steeped in Titan Lab context.
{{ msg.content }}
{{ error }}
The UI hits /api/ai/chat on the same host. External callers can hit the public endpoint
https://chat.ai.bstein.dev/api/ai/chat. POST JSON with a message and optional
history.
curl -X POST https://chat.ai.bstein.dev/api/ai/chat \
-H 'content-type: application/json' \
-d '{"message":"hi, what is your name?"}'
Responses stream in the UI. API callers get either a JSON body { reply, latency_ms } or a streamed text
body if available.