The Livepeer Engineering Stack

The key repositories, components, and layers that make up the Livepeer open source stack, and how they connect.

The Livepeer network is built entirely in the open. Every component – from the Go node implementation to the Solidity staking contracts to the AI inference runtime – lives in public repositories on GitHub under the livepeer organisation.

This page maps the main repos, explains how they connect, and points you to the right starting point for the kind of contribution you want to make.

Repo map

Repo	What it does	Language	Role in the stack
livepeer/go-livepeer	The Livepeer node – implements Broadcaster, Orchestrator, Transcoder, and Gateway roles	Go	Core runtime; everything else depends on or connects to this
livepeer/livepeer-protocol	Solidity smart contracts governing staking, bonding, and on-chain job settlement on Arbitrum	Solidity	Protocol layer; go-livepeer calls these contracts for on-chain operations
livepeer/ai-worker	The AI runner – containerised Python inference service for batch and real-time AI pipelines	Python	go-livepeer spawns ai-runner Docker containers per pipeline to handle inference
livepeer/comfystream	Real-time AI video pipeline engine built on ComfyUI; the primary tool for building live-video-to-video workflows	Python	Connects to the Livepeer network via the trickle protocol (see pytrickle)
livepeer/pytrickle	Python SDK for the trickle streaming protocol – the transport layer used by comfystream and BYOC pipelines	Python	Transport layer; used by comfystream and by custom BYOC containers (via PyTrickle FrameProcessor)
livepeer/livepeer.js / livepeer/ui-kit	Frontend SDK and React components for integrating Livepeer video into web applications	TypeScript / React	Application layer; independent of the AI and protocol stack
livepeer/docs	This documentation site	MDX (Mintlify)	Docs; you are reading the output of this repo right now

How the repos connect

go-livepeer is the foundation. It implements the node software that every participant in the network runs – in Gateway mode, Orchestrator mode, or both. When AI inference is enabled, go-livepeer spawns one or more ai-runner Docker containers (from livepeer/ai-worker) per GPU, and proxies inference requests to them via a local REST interface.

livepeer-protocol defines the on-chain rules. The Solidity contracts on Arbitrum govern staking, bonding, reward distribution, and slashing. go-livepeer calls these contracts when a node registers on-chain or settles payments. Off-chain AI gateways bypass this layer entirely – they talk directly to orchestrators without on-chain settlement.

comfystream sits above go-livepeer for the real-time AI path. It runs alongside ComfyUI and uses pytrickle to exchange frames with an orchestrator over the trickle streaming protocol. A BYOC developer using pytrickle directly implements their own FrameProcessor class – the same interface that comfystream uses internally.

livepeer.js / ui-kit is independent of the above. It is the application SDK for developers building Studio-integrated video apps – stream playback, upload, asset management. It does not interact with go-livepeer directly.

Where to start contributing

If you want to…	Start with	Look for
Build or improve AI pipelines (batch inference, new pipeline types)	livepeer/ai-worker	Open issues and runner/pipelines/ Python modules
Build real-time AI workflows (ComfyStream, BYOC, trickle transport)	livepeer/comfystream or livepeer/pytrickle	Open issues; ask in #comfystream Discord
Contribute to the core node (gateway, orchestrator, transcoder logic)	livepeer/go-livepeer	”Contributing to go-livepeer” guide in the README; open issues
Work on protocol contracts (staking, bonding, governance)	livepeer/livepeer-protocol	Open issues; Protocol R&D SPE (Sidestream) for mentorship
Build the frontend SDK (player, React components, Studio API integration)	livepeer/ui-kit	Open issues; #developers Discord
Improve documentation	livepeer/docs – branch docs-v2	Open issues tagged docs; contribution guide

All repositories accept pull requests. The go-livepeer README links directly to a contributing guide. For ai-worker and comfystream, open issues are the best starting point – ask in the relevant Discord channel (#comfystream, #developers, #protocol-development) before starting work on anything substantial.