Most agent frameworks assume the cloud: your keys, your data, and your conversations all flow through someone else's servers. ElizaOS flips that default — it treats the agent as something you boot and own, running on your own device with the model, memory, and wallet kept local. The bet is that an agent's value comes less from one clever model and more from being a persistent, programmable presence across the apps you already use.
What Sets It Apart
- Model-agnostic core — swap between OpenAI, Anthropic, Gemini, Grok, Llama, or the on-device Eliza-1 family without rewriting agent logic, so a working agent can move fully offline with no code changes.
- Composable primitives — actions, providers, services, and evaluators mean behavior is assembled from plugins rather than hardcoded; one package can add voice, browser automation, document RAG, or an EVM/Solana wallet.
- Local-first and bootable — it ships as full Linux and Android distributions, not just a library, so an autonomous agent can run as the operating system itself with privacy as the default.
- Cross-platform reach — native connectors for Discord, Telegram, X, Slack, WhatsApp, iMessage, and Farcaster let one agent identity persist across the surfaces people actually message on.
Who It's For
Great fit if you want an agent you control end to end — privacy-sensitive assistants, crypto-native automation, or experiments that must run offline on consumer hardware. Look elsewhere if you want a managed, point-and-click agent builder: the TypeScript plugin model, multi-platform wiring, and on-device model setup assume you are comfortable with code and infrastructure, and the wallet and on-chain features add surface area many projects will not need.
