Omnara

Omnara is an open-source platform for deploying and running AI agents in production, positioned as a self-hostable alternative to Claude Managed Agents and OpenAI's Agents API. You describe an agent in a short YAML config, launch it with a single CLI command or the TypeScript SDK, and Omnara handles execution, state and streaming while you choose the model, tools and machine.

How it works. Agent state lives in a durable event log, so a run can last for days, recover after a failure and consume no compute while it sits idle. Agents can run purely on Omnara when they only need web search, fetch, custom tools or MCP servers, or attach a shell on Omnara-provided sandboxes, your own Blaxel, Daytona or Unikraft Cloud account, or a macOS or Linux machine running the Omnara daemon, which connects outbound so no ports need opening.

Model-agnostic. Works with OpenAI, Anthropic, OpenRouter, LiteLLM or any compatible endpoint, so open models such as Qwen, GLM and DeepSeek are an option too. Agent configs can be changed mid-conversation without redeploying.

Built for teams. Human approval for risky tool calls, encrypted and scoped secrets, role-based access for people and API keys, and an append-only audit history. Integrations include MCP servers, reusable skills and first-party Slack, Discord and GitHub connections.

Pricing and licence. The code is Apache 2.0 and free to self-host. On Omnara Cloud there is no platform fee: bring your own model keys and machines, or pay as you go with credits at provider token rates plus active machine time. Docs and quickstart are at docs.omnara.com; the TypeScript SDK is @omnara/sdk on npm.