ClaudeMap

Omnara

The open-source managed-agent platform — run agents on any model and any machine through one API, with an MCP server ready for Claude Code

Plugins & Agentsagent platformmanaged agentsmcpself-hostedopen source

Omnara is the open-source, self-hostable take on a managed-agent platform: define agents in YAML, launch them from the CLI, SDK, REST API, Slack, Discord, GitHub events or cron, and bring your own keys for OpenAI, Anthropic, OpenRouter, Bedrock, LiteLLM, Ollama or self-hosted endpoints. On the machine side, install the daemon on any macOS or Linux laptop, VM or container (no inbound ports, no SSH keys) or pick sandboxes from twelve providers — an agent can move between machines when one goes offline and sync its work back with git. Every step is saved, so agents resume after crashes and disconnects. Permissions are per tool — run immediately, ask a person first, or block — and Slack/Discord mentions start agents whose approvals happen in the thread, while GitHub triggers read diffs, comment and submit reviews. An MCP server connects Claude Code, Codex, Cursor or any MCP client, and the site ships a SKILL.md so your coding agent can wire the whole thing up itself. Self-host free under Apache 2.0, or use Omnara Cloud.

What it does

  • Any model via your own keys — OpenAI, Anthropic, OpenRouter, Bedrock, LiteLLM, Ollama or a self-hosted endpoint — defined per agent in YAML
  • Machines: the daemon runs on any macOS or Linux box with no inbound ports or SSH keys, or choose sandboxes from twelve named providers; agents hop machines and git-sync their work back
  • Durable state — every step saved, so agents pick up after crashes and machine disconnects; subagents are full agents you can inspect
  • Per-tool permissions: run immediately, ask a person first, or block — the platform enforces it and the agent waits for the approval
  • MCP server for Claude Code, Codex, Cursor and any MCP client, plus a hosted SKILL.md that lets your coding agent do the setup
  • REST API, TypeScript SDK (@omnara/sdk) and CLI (omnara); stream every event or receive webhooks, and message agents mid-run

When to use it

Developers who want Claude Code-style agents running unattended — on a schedule, on another machine, or out in the field — while approvals and steering happen from Slack, Discord or the dashboard. Also fits teams replacing ad-hoc cron scripts and SSH hops with one audited agent runtime.

Getting started

# requires Node.js 22+; the first launch opens a browser to sign in at app.omnara.com

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

# chat with the running agent
npx omnara agents chat <agent-id>

Things to know

3.4k stars as of writing (2026-11-06). Apache 2.0; self-host free or use Omnara Cloud at app.omnara.com. The quickstart's omnara-openrouter provider and default-pool sandbox pool come provisioned with your first project so the sample config runs as is — model usage still bills to the provider behind it (OpenRouter). MCP and SKILL.md setup paths need a reachable Omnara instance (cloud or self-hosted). The README positions Omnara against Claude Managed Agents directly: pick the managed service if you want one vendor's models as a service, Omnara if you want any model on your own machines.

Repository

omnara-ai/omnara

Charted

Related on the map

  • clawd-conduit

    Plugin / Agent

    Wire all 15 Claude Code hook events into the Clawd on Desk desktop pet — and turn permission requests into a blocking desktop card.

    • hooks
    • desktop
    • notifications
    • claude-code
  • CometixCode

    Plugin / Agent

    An unofficial Rust 1:1 reimplementation of the Claude Code terminal UI, ported file-for-file onto an iocraft TUI framework.

    • terminal-ui
    • rust
    • reimplementation
    • unofficial
  • Abide

    Plugin / Agent

    Hook-level enforcement for the natural-language rules in AGENTS.md and CLAUDE.md — every edit is judged and violations are repaired in the same turn.

    • hooks
    • code-review
    • rules
    • agents-md