Eliza integration

The Livepeer model provider plugin for the Eliza agent framework (ai16z). Route Eliza’s LLM inference through the Livepeer network’s decentralised GPU infrastructure.

Eliza is an open-source multi-agent framework from ai16z. It supports pluggable model providers: OpenAI, Anthropic, Google, and Livepeer. The Livepeer plugin (plugin-livepeer) routes all LLM inference through a Livepeer AI gateway, where requests execute on decentralised GPU infrastructure.

The plugin was built by the Agent SPE (approved April 2025 with 30,000 LPT) as part of the first production AI agent pipeline on Livepeer.

Minimal setup

{
  "name": "MyAgent",
  "modelProvider": "livepeer",
  "settings": {
    "model": "meta-llama/Meta-Llama-3.1-8B-Instruct"
  },
  "clients": [],
  "system": "You are a helpful AI assistant.",
  "bio": ["AI assistant on decentralised compute"],
  "lore": [],
  "knowledge": [],
  "topics": ["general"],
  "adjectives": ["helpful"]
}

# .env
LIVEPEER_GATEWAY_URL=https://dream-gateway.livepeer.cloud
# API key optional for community gateway; required for production gateways
# LIVEPEER_API_KEY=your-key

pnpm start --characters="characters/my-agent.character.json"

The modelProvider: "livepeer" field tells Eliza to load the Livepeer plugin and route all generate() calls through LIVEPEER_GATEWAY_URL. The model field accepts any Ollama-compatible model ID supported by the Livepeer LLM pipeline.

Plugin architecture

The Livepeer plugin implements Eliza’s IModelProvider interface. When Eliza calls provider.generate(), the plugin:

1. Formats the messages as an OpenAI-compatible chat completion request
2. POSTs to {LIVEPEER_GATEWAY_URL}/llm with the model ID and messages
3. Returns the response text to Eliza’s agent loop

The plugin does not modify Eliza’s memory, knowledge, or action systems. It replaces only the inference backend. All other Eliza features (RAG, actions, evaluators, client connectors) work unchanged.

Supported models

Any Ollama-compatible model on the Livepeer LLM pipeline works. Warm models respond immediately; cold models load on first request (30-90 seconds).

Model	Warm	VRAM
meta-llama/Meta-Llama-3.1-8B-Instruct	Yes	8 GB
mistralai/Mistral-7B-Instruct-v0.3	Check network	8 GB
google/gemma-2-9b-it	Check network	10 GB
Qwen/Qwen2.5-7B-Instruct	Check network	8 GB

Extending the agent

Client connectors: Add "slack", "discord", or "twitter" to the clients array. Configure credentials in .env. The Livepeer inference backend works with all Eliza client connectors.

Knowledge and RAG: Add entries to the knowledge array in the character file. Eliza’s built-in vector store indexes them for retrieval-augmented generation.

Multi-agent: Eliza supports running multiple agents in one process, each with different character files. Each agent can use a different modelProvider or the same Livepeer backend with different model IDs.

The AI agent on Livepeer tutorial walks through the complete setup from clone to running agent. The Eliza Livepeer plugin tutorial covers the plugin architecture in depth.