# Open Machine ### One MCP. Every machine. Real execution. **An open-source runtime that lets your existing AI agent operate your machines.**

Open Machine — Your AI. Your machines. Connect any AI to your context, and get things done on your machines.

https://github.com/user-attachments/assets/f22519e6-e3ad-405c-a725-9ce53a642bcc **Open Machine (OM Core)** connects your existing AI Agents to your machines through a local-first MCP runtime. Your AI plans the work; OM provides permission-controlled tools to carry it out and return results. [Documentation](https://openmachinelabs.github.io/openmachine/) · [Getting started](docs/getting-started/installation.md) · [GitHub Issues](https://github.com/OpenMachineLabs/openmachine/issues) · [Contributing](CONTRIBUTING.md) · [Code of Conduct](CODE_OF_CONDUCT.md)

Join the Open Machine Community
_{Discord · Q&A · Show and Tell}

## What you can do | Capability | Guide | | --- | --- | | Local machines and context | [Inspect the intended project and environment](docs/features/local-machine.md) before acting. | | Workspace files and worktrees | [Read, search and apply guarded edits](docs/features/workspace.md) in the intended tree. | | Native commands and terminals | [Run commands, retrieve original results and verify cleanup](docs/features/execution.md). | | Remote Linux over SSH | [Prepare workers, pin identity and operate remote projects](docs/features/ssh.md). | | Persistent tasks | [Keep goals, stages and evidence; continue original work explicitly](docs/features/tasks.md). | | GitHub | [Use scoped repositories, Issues, PRs, reviews, Projects and Releases](docs/features/github.md). | | Browser subtasks | [Delegate a bounded web task](docs/features/browser.md) through a compatible, authorized native Codex installation. | | Skills | [Discover packaged operating guides and reviewed external Skill roots](docs/features/skills.md). | | Service and runtime | [Manage the macOS service, retained updates and compatible rollback](docs/features/service.md). | **Long-running jobs and task continuation:** Already-started jobs can keep running after a ChatGPT conversation disconnects. A job's execution status (running, completed, or failed) is separate from whether ChatGPT is ready to continue or needs user input. Finishing a job **does not automatically wake ChatGPT or start the next stage**. For a structured task, use the original [Task ID](docs/features/tasks.md) in a new conversation to review recorded results, failures, and pending approvals, then explicitly continue the remaining authorized work without rerunning completed steps. Optional integrations have separate prerequisites. See [current limitations](docs/introduction/limitations.md). ## Quick start **macOS Apple Silicon (macOS 13+, Node.js 24 LTS recommended; Node.js 22 LTS also supported)** ```bash npm install -g openmachine@latest openmachine ``` On first run, choose an existing project folder containing a readable file (such as `README.md`). Open Machine sets up Core and starts a managed macOS user service that keeps running after you close the terminal. Check it with `openmachine status`. Optionally enable startup at login: ```bash openmachine startup on ``` The npm-managed route is the default and only supported consumer installer today. The package currently targets Apple Silicon macOS; Linux and other source-based setups remain developer/advanced paths until npm packages for those platforms are published. Windows is not supported. To use ChatGPT, [connect it separately](docs/getting-started/connect-chatgpt.md) and [verify the connection](docs/getting-started/verification.md). The npm installer does not set up an MCP tunnel or approve client tools. Command execution and workspace writes are **disabled by default**. The npm `openmachine` command manages setup and service status; `openmachine core ...` runs the underlying Python CLI. See the [npm setup guide](docs/getting-started/npm-macos.md). This is an early-development release, not a stable 1.x release. **Updating?** Use the same npm-managed path for both launcher and running Core. See [Updates](docs/getting-started/updates.md). ```bash npm install -g openmachine@latest openmachine update ``` ## How it works

You request a task in an AI Agent, which plans and uses OM Core to perform authorized actions on your machines. Results return to the AI Agent. Gmail, Slack, GitHub, and Personal Context are shown as context sources above OM, with context updates flowing back.

Your request flows from **AI Agent → OM Core → your machines**, and execution results return to your AI. OM exposes the approved tools; it does not replace your AI agent. The **Personal Context layer and automatic context updates** in the diagram illustrate planned functionality for bringing data from sources such as Gmail and Slack into OM and keeping it up to date as tasks are completed. ## Documentation The [user documentation](https://openmachinelabs.github.io/openmachine/) covers every important released Core capability. Start with [requirements](docs/introduction/requirements.md), [installation](docs/getting-started/installation.md), [configuration](docs/getting-started/configuration.md) and [layer-by-layer verification](docs/getting-started/verification.md). For existing installations, use [updates](docs/getting-started/updates.md), the [configuration reference](docs/configuration/reference.md) and [troubleshooting](docs/troubleshooting.md). The [CLI](docs/reference/cli.md) and [MCP tools](docs/reference/tools.md) references link to detailed operating contracts without duplicating them. The [operating Skill](openmachine/skills/overview/SKILL.md) and [Skills directory](openmachine/skills/README.md) remain the technical starting points for AI agents. A source checkout, published release, running runtime and approved client catalog can differ; read the running version's guide before acting. ## Roadmap The [current development roadmap](ROADMAP.md) outlines planned features and longer-term directions. Scope and priorities may change. ## Security Open Machine is a trusted, single-operator native runtime, **not an OS sandbox** or multi-tenant executor. Native commands can access resources beyond read-only tool scopes. Keep the listener private, grant access deliberately, and remember that files and logs returned to an AI client may leave your machine. Cancellation and software rollback do not undo completed effects. Read [Security and data](docs/configuration/security.md) before enabling execution or remote access; never put secrets in prompts, command arguments or committed configuration. Report vulnerabilities privately via the [Security Policy](SECURITY.md). ## Contributors Thanks to everyone who contributes to Open Machine.

[View GitHub contributors](https://github.com/OpenMachineLabs/openmachine/graphs/contributors) · [Contribute to Open Machine](CONTRIBUTING.md) · [Maintainer guide](MAINTAINERS.md) ## License [MIT](LICENSE) © 2026 OpenMachineLabs.