Dev Resources
Omnara favicon

Omnara

AI/Programming/Tooling

Command Center for AI Coding Agents.

Omnara's website screenshot

About Omnara

Omnara is an API and platform for building and running production-grade AI agents. It handles hosting, durable execution, and state management so that agents can run for extended periods, survive crashes, and resume from where they left off, with every turn, tool call, and approval persisted as durable state rather than in-process memory. It works with any model, including OpenAI, Anthropic, OpenRouter, and self-hosted or compatible endpoints, and provides both managed pool machines that agents can create and destroy on demand, and support for connecting a team's own devices and environments.

The platform includes role-based access control, scoped and encrypted secrets, and enforced tool policies (always_allow, always_ask, always_deny) applied at the platform level rather than the prompt level. It maintains an append-only, database-enforced audit history of every tool call and event, and supports approval workflows for gating risky actions such as production deployments. Additional features include resumable streaming (SSE), machine pools, checkpoints for snapshotting agent state, durable artifacts, MCP integration, and versioned reusable "skills."

Omnara is open source and self-hostable, with an agent's state represented as a queryable, exportable event log rather than opaque vendor-held data. It is intended for teams and companies building their own production agent infrastructure - such as internal coding agents, cloud developer agents, or customer-facing automation - who would otherwise need to build durability, access control, and observability infrastructure themselves. The product is in early access with a waitlist for sign-up.

Key features

  • Durable agents that survive crashes and resume exactly where they left off
  • Managed pool machines created on demand, or connect your own devices and environments
  • Role-based access control with scoped credentials and encrypted secrets
  • Approval gates for risky actions with an append-only, queryable audit history
  • Works with any model, including OpenAI, Anthropic, OpenRouter, or self-hosted endpoints
  • Open source and self-hostable platform

Frequently asked questions

Which models does Omnara support?

Omnara works with OpenAI, Anthropic, OpenRouter, or any compatible endpoint, including models you host yourself.

Is Omnara open source?

Yes, the whole platform is open source and can be self-hosted.

Do I need to run my own infrastructure?

No — you define the agent's model, machines, tools, and permissions, while Omnara owns hosting, state, and durable execution.

Who owns the agent's data?

An agent's data lives in an event log rather than opaque vendor state, so you can query, export, and own it.

How does Omnara keep humans in control of agent actions?

Tool policies (always_allow, always_ask, always_deny) are enforced by the platform, and risky actions require approval that is recorded in an append-only audit history.

Advertise here

Featured products

  • SerpApi - Search API favicon
  • CoreClaw favicon

Show your product to thousands of developers

· 100k monthly pageviews
· 7k newsletter subscribers

Advertise your product