elizaOS/eliza

elizaOS – an open‑source, TypeScript‑based operating system for autonomous AI agents

What it is

• A monorepo that bundles the runtime, CLI, web/desktop/mobile app, cloud services, native bridges, and first‑party plugins needed to run Eliza, an agentic personal‑assistant platform.
• The core is a model‑agnostic framework (@elizaos/core) that defines the agent runtime, message loop, memory primitives and a plugin system. Plug‑ins add capabilities such as chat, voice, document handling, calendar integration, device sensors, blockchain wallets, and on‑device inference.
• The project ships the Eliza app (web, desktop, Android/iOS) that users interact with, and a beta CLI (elizaos) for scaffolding new agents or plugins.
• A separate repository (elizaOS/os) contains bootable Linux and Android images that can run the whole stack as a dedicated OS.

Key components

Package	Role
@elizaos/core	Defines AgentRuntime, message handling, memory/state, and plugin contracts.
@elizaos/agent	Wraps the core into a standalone HTTP‑backed agent service.
@elizaos/app-core	Shared hosting logic for the Eliza UI across platforms.
@elizaos/ui	React UI library used by the web/desktop/mobile shells.
elizaos (CLI)	Generates new projects or plugins, handles upgrades and deployment.
plugins/	First‑party plugins: local inference, connectors (e.g., calendar, email), device bridges (camera, contacts), blockchain wallets, etc.
@elizaos/plugin-local‑inference	Enables on‑device execution of the Eliza‑1 models (2 B‑27 B parameters, Gemma‑based) plus local embeddings, speech, vision and image‑generation assets.
Eliza Cloud (optional)	Hosted services for authentication, model routing, remote connectivity and cross‑device sync.

Local vs. cloud inference

• The runtime can route each capability to local (downloaded GGUF models), a direct provider (e.g., OpenAI, Anthropic), or Eliza Cloud. Hardware detection automatically picks a supported backend; if the device cannot run the model, it falls back to a remote provider.

Getting started

1. Run from source (requires Bun ≥ 1.x and Node):

   git clone --filter=blob:none https://github.com/elizaOS/eliza.git
   cd eliza
   bun install          # installs deps, submodules, patches, and builds libelizainference
   bun run dev          # starts the development server
   

   • bun run fetch:archive-artifacts pulls large model assets if needed.
   • Common scripts: bun run build, bun run test, bun run cloud:mock.
2. Use the CLI to scaffold a new project or plugin:

   bun add --global elizaos@beta
   elizaos create my-project --template project
   elizaos create plugin-example --template plugin
   

3. Try the hosted product – open https://cloud.eliza.app or download a pre‑built release from the GitHub releases page.

Typical workflow for developers

• Write a plugin that exports a Plugin object (register actions, providers, routes, UI views, etc.).
• Run bun run test:e2e to verify integration against a real runtime.
• Follow the evidence‑based review process described in AGENTS.md before merging.

Target audiences

• Product teams building AI‑powered personal assistants, automation bots, or on‑device agents.
• Developers who want a plug‑in architecture to experiment with different LLMs, embeddings, or multimodal models without being locked to a single provider.
• Researchers interested in on‑device inference of modern LLMs (Gemma‑based) and multimodal assets.

License

• MIT – permissive, allowing commercial use, modification, and redistribution.

---

Bottom line: elizaOS is a full‑stack, open‑source platform that lets you run autonomous AI agents locally or in the cloud, extend them via a rich plugin system, and even ship a dedicated OS image that turns a device into a self‑contained AI assistant.