← Back to feed

Omnara has published an open-source managed agent platform at github.com/omnara-ai/omnara under Apache 2.0. It positions itself as a self-hostable replacement for both Claude Managed Agents and OpenAI’s Agents API — the two dominant proprietary agent hosting services from Anthropic and OpenAI respectively.

The project hit Hacker News’ front page today.

What It Does

Omnara handles the infrastructure layer agents need: tool access, secrets management, permissions, sandboxed execution environments, and streaming output. Agents are defined in YAML and run via the CLI or TypeScript SDK.

A minimal agent config:

version: v1
instruction: |
  You are an engineering assistant. Use your machine to run commands.
model:
  provider_config: omnara-openrouter
  name: anthropic/claude-fable-5.1
machine_sources:
  - machine_pool_name: default-pool

Launch via CLI:

npx omnara agents launch --file agent.yaml \
  --name "Machine assistant" \
  --message "Report this machine's operating system and free disk space."

The provider_config field routes to any model through OpenRouter or custom provider integrations, decoupling agent logic from model vendor lock-in. Agents can use Claude Fable 5.1, GPT-6 Astra, or any other OpenRouter-available model by changing a single line.

Self-Host or Cloud

Self-hosting requires Docker with Compose and a git clone followed by docker compose --profile app up -d. The development setup prints verification links in Docker logs rather than requiring an email provider.

Omnara Cloud at app.omnara.com is available for teams that don’t want to operate infrastructure. New projects get a default provider config pointing to OpenRouter and a pool of managed sandboxes out of the box.

TypeScript SDK

The SDK (@omnara/sdk on npm) wraps the REST API for programmatic agent lifecycle management. A launched agent returns an agent.id; callers subscribe to a streaming event channel to observe tool calls, intermediate steps, and the final response. The pattern mirrors OpenAI’s streaming Responses API but adds agent persistence and tool sandboxing as first-class primitives.

The Competitive Context

Claude Managed Agents launched in April 2026 at $0.08/hour per running agent. OpenAI’s Agents API has been in public beta since September 2026 at token-only pricing with no orchestration fee. Both require routing agent workloads through the respective vendor’s cloud.

Omnara’s pitch is control: run agents on your own infrastructure, against any model, with secrets that never leave your environment. The Apache 2.0 license means enterprise teams can fork and modify without restriction. For organisations with data residency requirements or vendor diversification mandates, this is a workable path that wasn’t cleanly available until now.

The model-agnostic design is the sharper edge. Anthropic’s managed agents run Anthropic models; OpenAI’s Agents API runs OpenAI models. Omnara runs either — or DeepSeek, Gemini, or anything else accessible through OpenRouter.