Omnara is a managed agent platform: an API that launches AI agents with a model, tools, secrets, permissions, and machines defined in a YAML config, then streams their events back. It runs as Omnara Cloud or self-hosted, and positions itself as an open-source counterpart to Claude Managed Agents and OpenAI's Agents API that works with any model.
Your app, the CLI, and the dashboard all use the same REST API to launch agents, send messages, and answer approvals. Postgres stores every step of every agent as an event log, and that log is the agent's state, so any worker can resume an agent after a crash or restart. Workers are stateless processes that build the model context, call the model, and run tools, and an agent holds a worker only while it takes a turn. Redis, blob storage, and a memory filesystem back the rest.
Machines are tools the agent uses. A daemon on a macOS or Linux laptop, VM, or container connects out to Omnara, so agents run commands there with no inbound ports or SSH keys. Agents can also use sandboxes from nine providers, including Daytona and Modal, use several machines at once, and continue in a sandbox when a laptop goes offline, syncing work back with git.
| Omnara | Claude Managed Agents | OpenAI's Agents API | |
|---|---|---|---|
| Models | Any compatible model | Claude | OpenAI |
| Where the platform runs | Omnara Cloud or your own infrastructure | Anthropic's cloud | OpenAI's cloud |
| Where agents run | Sandboxes from many providers, or your own machines | Anthropic sandboxes or your own | OpenAI or partner sandboxes, or your own machines |
| Agent history | Omnara Cloud, or your own Postgres when self-hosted | Stored by Anthropic | Stored by OpenAI |
Example agents include a Slack coding agent that opens pull requests, an SRE agent that investigates production problems, a browser agent, a PostHog analytics agent, social signal digests for Reddit, X, and LinkedIn, and a video generation agent. Each example is an agent config plus a skill file that a coding agent can deploy from one prompt.
Features
- Any model: bring your own keys for OpenAI, Anthropic, OpenRouter, LiteLLM, Ollama, or any compatible endpoint, including self-hosted models
- Tools: built-in shell, file, web search and fetch, and ask-a-person tools, plus MCP servers and custom tools that run in your own systems
- Skills: package instructions and files once, attach them to any agent, and the agent loads them when needed
- Memory stores: files that persist between conversations and can be shared across agents with read or read-write access
- Subagents: an agent starts subagents with fresh context, checks their progress, and sends them new instructions
- Machines and sandboxes: a daemon turns any laptop, VM, or container into a machine, or agents use sandboxes from nine providers
- Durable state: every step is saved, so agents resume after crashes, restarts, and machine disconnects
- Schedules: cron triggers with time zones message an agent, launch a new one, or start a Slack or Discord thread
- Slack, Discord, and GitHub: start agents from a mention or a pull request, with replies, questions, approvals, and code reviews in place
- Permissions: each tool runs immediately, asks a person first, or is blocked, enforced by the platform
- Multi-tenant access control: organizations, projects, roles for users and API keys, and shared secrets
- API, SDK, CLI, and MCP server: a REST API, a TypeScript SDK, a CLI, and a hosted MCP server for coding agents
