Omnara
Omnara is an open-source API for running agents in production. Teams choose their models, tools, and machines while Omnara handles execution and state. Its official page positions the service around agents embedded in product dashboards, internal Slack agents, code-review workflows, and interactive command-line experiences.
Launch and observe agents through an API
Developers can launch, steer, and observe agents through a REST API, TypeScript SDK, or CLI. Omnara describes agents as durable objects whose state survives failures. This makes it relevant to applications where a user needs to return to a continuing task, rather than start a new stateless request each time.
The platform supports models behind compatible endpoints, including providers such as OpenAI and Anthropic, routing services such as OpenRouter, and self-hosted models. Compatibility with the endpoint is the important condition; the provider examples do not mean every model exposes identical capabilities.
Choose machines and control actions
Omnara can manage machines, connect a team's own Blaxel, Daytona, or Unikraft Cloud account, or use a daemon installed on a macOS or Linux laptop, VM, or container. The daemon connects outbound, so it does not require opening inbound ports. The publisher also describes a hosted cloud and an option to self-host the platform.
Tool policies can pause an agent for human approval. Other listed controls include encrypted scoped secrets, role-based access for people and API keys, and an append-only action history. Teams can connect MCP servers and reusable skills, and use the first-party Slack integration to communicate with agents. These controls are useful design choices to examine when deciding which actions a deployed agent should be allowed to perform unattended.
Separate platform fees from resource costs
The pricing page lists a $0 platform fee and no per-seat fee for self-service access. Bringing your own model keys or machines is free at the platform level, and self-hosting is available under the Apache 2.0 license.
Omnara-managed models are billed at provider token rates. Managed machines have both runtime and retention charges: $0.0414 per GiB of memory per hour online and $0.20016 per GiB of memory per 30 days retained. Idle machines auto-sleep, pausing runtime charges while preserving data. Include retention as well as active use when estimating the cost of long-lived agents; free platform access does not make managed resources free.