Omnara, an open-source platform for running AI agents, pitches itself as a self-hostable alternative to Claude Managed Agents and OpenAI's Agents API

Omnara is an open-source platform for running AI agents, and it presents itself as a self-hostable alternative to Claude Managed Agents and OpenAI's Agents API.

According to the project, you use an API to run agents on any model and any machine. Tools, secrets, permissions and streaming are built in. You can self-host it for free under the Apache 2.0 license, or use the hosted Omnara Cloud.

Getting started on Omnara Cloud means signing in at app.omnara.com. The project says a first project comes with a model provider and a pool of managed sandboxes, which are isolated environments where agents run commands. You write a short agent.yaml config file, then launch it with a command-line tool that needs Node.js 22 or newer.

Self-hosting requires Docker with Compose. You clone the repository and start it with a Docker Compose command, then open it locally at port 8000. A new instance has no models or machines, so you add a model provider and connect a computer in the console. The project warns that the local defaults are intentionally insecure and points to its self-hosting guide for real deployments.

The design is fairly simple. Your app, the CLI and the dashboard all use one REST API. Every step of every agent is saved in Postgres, so another worker can pick an agent back up after a crash or restart. Workers are stateless processes that build the model context, call the model and run tools. Machines run a small daemon called omnarad that connects out to Omnara, so no inbound ports are needed. Omnara also uses Redis, blob storage and a memory filesystem.

On models, the project says you can bring your own keys for OpenAI, Anthropic, OpenRouter, Amazon Bedrock, LiteLLM, Ollama or any compatible endpoint. Its FAQ says any model behind an OpenAI Responses, OpenAI Chat Completions or Anthropic Messages compatible endpoint should work.

The feature list covers built-in tools for shell commands, files, web search and asking a person a question. It also supports MCP servers (a standard way to plug tools into AI apps) and custom tools. Other listed features include skills, persistent memory stores, subagents, cron schedules, and integrations with Slack, Discord and GitHub. Each tool can be set to run immediately, ask a person first, or be blocked.

For machines, agents can run on macOS or Linux laptops, VMs or containers. They can also use sandboxes from eleven providers: Arker, Blaxel, boxd, CreateOS, Daytona, E2B, Freestyle, Modal, Superserve, Tenki and Unikraft. The project says an agent can use several machines at once and can continue in a sandbox if a laptop goes offline.

Omnara also includes a comparison with the two hosted services. It says Claude Managed Agents is closed and runs on Claude models, while OpenAI's Agents API runs on OpenAI models. In the project's account, Claude Managed Agents costs Claude token rates plus $0.08 per active session-hour. It says OpenAI's option uses OpenAI model rates plus standard rates for hosted tools and sandboxes. These are the project's own claims and should be checked against the vendors' documentation.

On pricing, Omnara says self-hosting is free. It says Omnara Cloud has no platform fee: you can bring your own model keys and machines at no cost, or pay as you go for its models and sandboxes.

The repository also lists example agents. They include a Slack coding agent, a pull request review agent, an SRE agent for problems such as a 5xx error spike, and a browser agent. You can connect to Omnara from coding tools such as Claude Code, Codex and Cursor through an MCP server at app.omnara.com/mcp.