An OS for agents that isn't just a Python script

What it does elizaOS is a TypeScript framework for building autonomous AI agents and the applications that host them. It combines a model-agnostic runtime (@elizaos/core), a CLI for generating projects, a React-based management dashboard, and a plugin ecosystem spanning Discord, Telegram, Farcaster, browser automation, and RAG document ingestion. The repo enforces a strict split between the headless runtime, deployable product workspaces, and UI-contributing app plugins so the layers do not collapse into each other.

The interesting bit The architecture borrows from operating systems: a small core runtime, a plugin registry that acts like a package manager, and first-party app plugins that add entire surfaces such as voice telephony, browser control, or multi-agent task coordination. You can import just the runtime into your own TypeScript code, or scaffold a full Vite + React product with deployment scripts for Vercel, AWS, or Convex.

Key highlights

• Supports major LLM providers including OpenAI, Anthropic, Gemini, Llama, and Grok through a model-agnostic layer
• Ships 30+ independent examples covering chat, on-chain trading, MCP/A2A protocols, and serverless hosting targets
• Bundles evaluation suites for GAIA, SWE-bench, and tau-bench benchmarks
• Curated plugin registry plus first-party app plugins for phone, browser, documents, and form-driven data collection
• Runs a full local mock cloud stack with PGlite and control-plane for development without external cloud dependencies

Verdict Worth evaluating if you are building multi-agent products or need connectors across Discord, Telegram, and Farcaster; skip it if you are looking for a lightweight, single-purpose LLM wrapper.