Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.livepeer.org/llms.txt

Use this file to discover all available pages before exploring further.

Livepeer supports AI inference, real-time AI video, video transcoding, custom containerised models, and protocol-level contribution. These paths share infrastructure but have different APIs, SDKs, and architectural requirements.

​
Choose a build track

Path: AI Gateway APISend REST requests to a Livepeer gateway and receive processed outputs. No GPU required. The gateway handles routing to network orchestrators.Nine batch pipelines are available: text-to-image, image-to-image, image-to-video, image-to-text, audio-to-text, text-to-speech, upscale, segment-anything-2, and LLM.
curl -X POST https://dream-gateway.livepeer.cloud/text-to-image \
  -H "Content-Type: application/json" \
  -d '{"prompt": "a glowing neural network in a dark room", "model_id": "SG161222/RealVisXL_V4.0_Lightning"}'
The community gateway at dream-gateway.livepeer.cloud accepts unauthenticated requests for development. Production applications use a self-hosted gateway or a gateway provider with an API key.
Path: ComfyStreamComfyStream applies ComfyUI workflows to live video streams at up to 25 FPS on an NVIDIA RTX 4090. Build your pipeline as a ComfyUI workflow JSON file; ComfyStream handles WebRTC transport and frame processing.Requirements: NVIDIA GPU (RTX 3090 or better), ComfyUI, Python 3.12, CUDA 12.5+.Supported models: StreamDiffusion, ControlNet, IPAdapter, FaceID, LoRA, Whisper (audio), Gemma (video understanding), SuperResolution.
Path: BYOC (Bring Your Own Container)Package your model in Docker, implement PyTrickle’s FrameProcessor interface, and register the container as a worker on the Livepeer network. Your container receives and publishes live video frames over the trickle streaming protocol.Requirements: Docker on Linux with NVIDIA GPU, Python model code (PyTorch recommended).BYOC reached production-grade in Phase 4 (January 2026). Embody SPE and Streamplace run production BYOC workloads on the network.
Path: Video APIThe Livepeer REST API and SDKs manage the full lifecycle of livestreams and VOD assets: create streams, get ingest URLs, transcode assets, manage playback, and set access control.SDKs: npm install livepeer / pip install livepeer / go get github.com/livepeer/livepeer-go
Path: OSS contributionFour primary repositories: go-livepeer (protocol node, Go), ai-runner (AI inference runtime, Python), ComfyStream (real-time AI video, Python), protocol (Solidity contracts).Requirements: Go 1.21+ for go-livepeer; Python 3.12 for ComfyStream and PyTrickle; local testnet for integration testing.
Path: EvaluationUse the infrastructure stack concept page for the access layers, then the AI pipelines page to match your use case to a path.Key distinctions:
  • Batch vs real-time AI: different APIs, latency targets, and infrastructure requirements
  • Managed gateway vs self-hosted: managed is simpler; self-hosted controls cost at scale
  • Standard pipelines vs custom model: BYOC for custom architectures; gateway API for the nine supported pipeline types
then

​
Build path reference

Most developers start with the AI Gateway API path. It requires no GPU, no deposit, and produces a working inference call in under ten minutes. Move to ComfyStream or BYOC when you need real-time processing or a model architecture outside the nine standard pipelines.

Concepts

The developer stack, AI pipelines, video infrastructure, and the OSS codebase.

AI quickstart

First AI inference call in 10 minutes. No GPU required.

AI pipelines

All nine batch pipeline types with request schemas and curl examples.
Last modified on May 19, 2026