Two ways to contribute real LLM compute
Inference is a pluggable backend. Contribute by running a model locally, or by proxying requests to a provider API you hold a key for. Either way the job settles in native MATRIX through consensus, priced per unit and capped at the amount reserved up front.
Local runner
Serve inference from your own hardware, or from the built-in echo backend with no GPU at all.
- Ollama-style local HTTP backend for on-device models
- GPU-free echo backend for development and testing
- Earn native MATRIX for the tokens your machine actually serves
Provider-API proxy
Already pay for an OpenAI-compatible API? Contribute by proxying requests through it.
- OpenAI-compatible chat completions backend
- API key read from an environment variable, never stored in code
- Jobs settle in native MATRIX through consensus, priced per unit and capped at the reservation
Metered and settled through consensus
Jobs arrive over the matrix.inference.v1 InferenceService. A job is reported complete only once its settlement has committed, so a buyer is charged exactly once, for work that really happened, and never more than it reserved.
BUYER=$(matrix wallet show | awk '/^account:/{print $2}')
matrix fund --account "$BUYER" --amount 1000000
matrix provider register --id demo-inference-provider --capacity 1000 --price 5
matrix inference submit --buyer "$BUYER" \
--provider demo-inference-provider --prompt "hello world"Against a fresh node the buyer must be funded and the demo provider registered before a job can reserve and settle, so the block does both first. It assumes a local wallet (matrix wallet create) and a running node. Because a default node runs with ACLs enabled, add --api-key <key> to each command (the key the node was started with); on an open dev node with ACLs disabled you can drop it, though that node will not expose reward-pool funding.
Learn how inference settles
Compute marketplace
How inference jobs are priced, reserved, and settled alongside the rest of the compute market.
Architecture
Where the pluggable inference backends sit in the node and how settlement flows through consensus.
Contribute inference compute
Serve a local model or proxy a provider API, and earn native MATRIX for the work you do. Download a build or read the docs.