Documentation

Pied Piper

Pied Piper is open-source AI inference, hosted by the community. Idle GPUs — in browser tabs, extensions, and native nodes — pool together to serve models behind a single OpenAI-compatible API. People with spare compute earn credits; people who need inference spend them.

The shape of it

There are two sides to Pied Piper. Borrowers point any OpenAI client at /v1 and send requests. Hosts run a model on hardware they already own and get paid per request they serve. A dispatch layer in the middle matches requests to live hosts and settles credits between the two.

If you only want to use models, start with the Quickstart. If you want to earn by lending your GPU, jump to Hosting.

Your first request

Grab a key from Settings → Keys, then point curl (or any OpenAI SDK) at the Pied Piper base URL. The wire format is identical to OpenAI's — only the host and key change.

curl
1curl https://mesh-zeta-eight.vercel.app/v1/chat/completions \
2 -H "Authorization: Bearer $PIEDPIPER_API_KEY" \
3 -H "Content-Type: application/json" \
4 -d '{
5 "model": "llama-3.2-3b-instruct-q4f32_1",
6 "messages": [{ "role": "user", "content": "Hello from Pied Piper" }]
7 }'
Already using OpenAI?
Override baseURL to https://mesh-zeta-eight.vercel.app/v1 and swap in your Pied Piper key. Existing code keeps working — see the SDK page.

Explore the docs