Livepeer

Runner connects applications and agents to media workflows. Compute runs them.

Workflows define what happens. Runner makes them available through the CLI, MCP, and API. The Livepeer network supplies the compute that executes every run.

Three layers, one request

Runner

Runner is the interface between people, agents, applications, and Livepeer. It accepts a task, authenticates the caller, and starts the right workflow.

CLI · MCP with OAuth · API

Inference

Inference containers define the models, inputs, and runtime needed to produce a result on the network.

Select · run · measure usage

Compute

Orchestrators provide the GPUs that execute workflows. The network matches each run with available compute and returns the result to Runner.

Orchestrators · GPUs · rewards

How a Runner request moves

Intent

A person, agent, or application asks Runner to create or transform media.

Workflow

Runner selects the saved workflow and validates its inputs.

Execution

An orchestrator runs the workflow on its GPU and returns the output.

Result

Runner returns the output to the CLI, connected agent, or application.

Any agent can use Runner

Connect the Runner MCP and authorize it with OAuth. The agent gets access to this project’s workflows without handling a long-lived API credential.

Payment follows execution

The project pays when a workflow runs. The orchestrator that supplies the GPU earns service fees, while protocol rewards support active network compute.

Read the protocol architecture