An API to the GPU

Send a prompt; a GPU a few feet away runs a local model and sends the answer back.

⚡ Open the chat app → — a full chat with history and saved sign-in. This page is the API reference.

The Lab's GPU node accepts work over HTTP: you POST a request, it runs inference on a locally-hosted model, and returns the result — from the same machine that serves this page. It speaks a small native endpoint and an OpenAI-compatible one, so existing clients work unchanged.

Try it

Access is gated by a Matsya API key. Paste yours to run a prompt on the GPU. No key? Get one at matsyaai.com.

Endpoint

POST https://lab.knobly.com/gpu/infer
GET  https://lab.knobly.com/gpu/healthz                 (liveness, no auth)
POST https://lab.knobly.com/gpu/v1/chat/completions     (OpenAI-compatible)

Auth

Every call except /healthz authenticates with your Matsya API key (msk_…), sent as Authorization: Bearer <key>. Only valid Matsya keys are accepted — the GPU is not an open relay.

Example

curl -s https://lab.knobly.com/gpu/infer \
  -H "Authorization: Bearer $MATSYA_KEY" \
  -H "Content-Type: application/json" \
  -d '{"prompt":"Write a haiku about GPUs"}'