Skip to main content

The Livepeer GitHub organisation hosts over 170 public repositories. This page groups the ones developers actually touch by the network layer they belong to, and tags each one as [Official] (maintained under the livepeer/ organisation), [Community] (maintained by ecosystem contributors), or [Archived] (no longer maintained). For why each layer exists, see . For where each persona enters the stack, see .

Core Runtime Repositories

Every Livepeer Network role runs from one of these binaries. Every other layer in the stack depends on or connects to them. go-livepeer is the foundation. When AI inference is enabled, it spawns ai-runner Docker containers per GPU and proxies inference requests via a local REST interface. lpms is the media-processing library it uses for transcoding.

On-Chain Protocol Repositories

Cryptoeconomic contracts deployed on Arbitrum One. go-livepeer calls these for all on-chain operations. The confluence branch of livepeer/protocol is the deployed Arbitrum version. The streamflow branch covers the L1 contracts (LivepeerToken and BridgeMinter only; all other core protocol contracts are paused on L1).

Gateway API Surface

REST APIs and operator tooling that sit on top of go-livepeer.
The internal livepeer-data-mcp server is not a developer surface. It backs internal operations only.

Real-Time Pipeline Repositories

Live-video-to-video workflows, BYOC containers, trickle transport. ComfyStream sits above ai-runner on the real-time path and uses pytrickle to exchange frames with an Orchestrator. A BYOC developer using pytrickle directly implements the same FrameProcessor interface that ComfyStream uses internally.

Application SDK Repositories

Client libraries for calling the network from an application.
livepeer-js and ui-kit refer to the same active repository. The historical livepeer-js namespace redirects to ui-kit.

Agent and Creative Repositories

Agent SDKs and frameworks built on top of the network. Some packages are not yet published to npm. The two SDK packages live inside the storyboard monorepo until they ship on npm. Clone the repo to use them today.

Monitoring Repositories

Operator observability; mostly community-maintained.

Governance Repositories

Async governance and async-first coordination.

Documentation and Discovery Repositories

Solutions Repositories

Hosted products built on the network. Documented under ; listed here for completeness.

Layer Dependencies

The stack is a dependency chain. Every layer above depends on the layers below it; each layer can be accessed directly without building at the layers above. go-livepeer is the foundation. Every other official repository either depends on it, generates clients for its APIs, or provides tooling around it. ai-runner is spawned by go-livepeer per GPU per pipeline. go-livepeer generates its Go client bindings from ai-runner’s OpenAPI spec via make ai_worker_codegen. comfystream and pytrickle sit above ai-runner on the real-time path. comfystream uses pytrickle’s trickle protocol to exchange frames with an Orchestrator. A BYOC developer using pytrickle directly implements the same FrameProcessor interface. The AI Gateway API is the standard REST interface that all SDK consumers reach through. Calling livepeer-ai-js or livepeer-ai-python sends requests to a Gateway node, which routes them through go-livepeer to the relevant Orchestrators. ui-kit is independent of the AI and protocol path. It provides Player and Broadcast components for browser applications and works against any compatible Gateway.

Contribution Entry Points

covers the broader process.
Last modified on May 31, 2026