# omg.dev: A Local Control Plane for Parallel Coding Agents > omg.dev is an open-source harness that runs Claude Code, Codex, Cursor, and other coding agents on your own machine and lets you steer them from a single web UI. It trades built-in authentication for simplicity, so remote access requires a private tunnel like Tailscale. **BennyKok/omg.dev** — omg.dev — Remote control for claude, codex, cursor, opencode, pi, grok, jcocde with mobile client - Repository: https://github.com/BennyKok/omg.dev - Website: https://omg.dev - Stars: 549 · Forks: 42 - Language: TypeScript - License: MIT - Published: 2026-08-23 · Updated: 2026-08-23 · Language: en - Canonical page: https://hysenlabs.com/projects/bennykok-omg-dev ## What omg.dev Actually Solves The problem omg.dev addresses is the friction of running multiple coding agents side by side. Most agent CLIs are single-session tools: you open a terminal, start Claude Code or Codex, and you are tied to that one conversation. omg.dev turns your computer into a control plane that can run several agent sessions in parallel, and it gives you a web UI to read transcripts and send follow-up instructions. The intended user is a developer who already pays for agent subscriptions or has API keys, and who wants to supervise several autonomous coding tasks without juggling multiple terminal windows. The README is explicit that you use your existing agent subscriptions or API keys, so this is not a replacement for the agents themselves. It is a management layer on top of them. ## The Architecture: Local Control Plane, Web UI, Optional Hosted Version The repository describes a local control plane that runs on your computer. You install the CLI with Bun, run `omg computer setup`, and the server listens on port 8766. The web UI talks to that local server. The README mentions a hosted Computer option that runs omg.dev in the cloud and opens from a browser, but the local setup is the core. The desktop app for macOS uses Electrobun with a Bun main process, which is an unusual choice: it is a macOS-first shell that still depends on the local control plane. The data flow is straightforward: the control plane connects to installed coding agents, manages their sessions, and exposes transcripts and input controls through the web UI. The README also mentions Chat, Bots, Schedules, and Notifications as features in one place, which suggests the control plane does more than just proxy terminal output, but the details of how schedules or bots work are not in the provided material. ## Getting It Running: Commands and Config The install path is short. You need Bun installed first. Then you run `bun install --global @omg-dev/cli && omg computer setup`. After that, you open `http://localhost:8766`. The setup supports Debian or Ubuntu Linux and macOS. On Linux, you must run it as a normal user with `sudo` access, and the README explicitly warns not to run it as `root`. That is a concrete constraint that many local tools ignore. For remote access from a phone, the README gives a specific command: `OMG_TAILSCALE_SERVE=1 omg setup`. That environment variable is the only config key mentioned. Management commands are also documented: `omg computer status`, `omg computer update`, `omg doctor`, and `omg computer uninstall`. The uninstall command keeps sessions and config, which is a nice touch for a tool that manages long-running agent tasks. There is no mention of a configuration file for the server itself, so most behavior is controlled by environment variables and the setup wizard. ## Security Model: No Built-In Auth, Tailscale Is the Answer The most important limitation is in the security section. The local server binds to `127.0.0.1` by default and has no built-in authentication. The README says plainly: do not expose it directly to the public internet. That is a significant gap for a tool that manages coding agents, which can execute arbitrary commands on your machine. The recommended workaround is to serve it privately through Tailscale, using the `OMG_TAILSCALE_SERVE=1` flag. This is a reasonable pattern for a personal tool, but it means you must trust your Tailscale network and your own configuration. There is no mention of API tokens, user accounts, or per-session permissions. If you want to share access with a team, you would need to build your own authentication layer in front of the local server, which is not documented. The README points to SECURITY.md and docs/remote-access.md, but those files are not in the provided material, so the full security story is incomplete. ## Supported Agents and the macOS Desktop Preview The control plane supports a specific list of agents: Claude Code, Codex, Grok, Cursor, fx, OpenCode, Jcode, GitHub Copilot, and Pi. That is a broad set, covering both major commercial agents and open-source ones like OpenCode. The README says you install or connect an agent from the Settings page, and you authenticate with the agent provider. That means omg.dev does not handle authentication itself; it relies on each agent's existing login flow. For the desktop app, there is a macOS preview that only supports Apple Silicon. It is not signed or notarized, so you have to go through the Control-click and Open flow, or use System Settings to allow it. That is a real friction point for non-technical users, but for the target audience of developers, it is acceptable. The desktop app is not a standalone product; it uses the local control plane, so you must run the install command first. ## Limitations and Failure Modes The most obvious limitation is the lack of authentication, which I already covered. Another limitation is the platform support: only Debian or Ubuntu Linux and macOS. Windows users are out of luck, and there is no mention of other Linux distributions. The macOS desktop preview is not signed, which can be a blocker in managed corporate environments. The README also warns against running as root, which is a common mistake that could lead to permission issues or security problems. A failure mode that is not documented but is implied: if you expose the server to the internet without Tailscale, you are giving anyone who finds the port full control over your coding agents, which means arbitrary code execution on your machine. That is a severe risk. Also, because the server has no built-in authentication, any process on your local machine that can reach port 8766 can control your agents. On a multi-user system, that is a concern. The README does not mention any rate limiting or session isolation, so the assumption is that this is a single-user tool. ## Alternatives and How They Differ A common alternative is to run each coding agent in its own terminal or IDE tab. That approach has no central UI, no transcript aggregation, and no way to send follow-up instructions from a phone. omg.dev's value is the aggregation and the remote access. Another alternative is a hosted service that runs agents in the cloud, such as the hosted Computer option that omg.dev itself offers. The difference is where the agents run: omg.dev's local mode keeps your code and your agent sessions on your machine, which matters if you have privacy constraints or large repositories that are slow to upload. The hosted version runs in the cloud, which removes the need to install anything but introduces the question of where your code goes. A third alternative is to use Tailscale's own serve feature to expose a single agent's CLI, but that does not give you the multi-agent management or the web UI. omg.dev's approach is to be the layer that ties multiple agents together, not to replace any single agent. ## Maintenance and Upgrade Costs The project is under active development, with three releases in the span of two days in the provided data (v0.6.15, v0.6.16, v0.6.17). That suggests a fast iteration cycle, which is good for bug fixes but means you should expect frequent updates. The README provides a simple update command: `omg computer update`. The uninstall command preserves sessions and config, which reduces the cost of moving to a new version or rolling back. The license is MIT, which is permissive and imposes no obligations beyond preserving the copyright notice. There is no mention of a migration guide or a changelog in the README, but the release tags imply that you can inspect differences between versions. The maintenance cost is low for a single user, but if you rely on the hosted version, you are dependent on the project's uptime and its pricing, which is not documented. For a local setup, your main cost is keeping the server updated and ensuring your agent provider's authentication does not break when you update the CLI. ## Conclusion Adopt omg.dev if you are a developer on Debian, Ubuntu, or macOS who wants to run multiple coding-agent sessions in parallel and manage them from a browser or phone, and you are comfortable with a local server that has no built-in authentication. Do not adopt it if you need fine-grained access control out of the box, if you run Windows, or if you require a signed desktop app for macOS. Before using it, verify that your agent provider's authentication works with the CLI, and read SECURITY.md and docs/remote-access.md to understand how to expose the server safely, preferably through Tailscale. The project is under active development with frequent releases, so check the changelog for breaking changes before upgrading. ## FAQ ### What is omg.dev? It is an open-source parallel coding agent harness. Agents run on your own computer, you control them from one web UI, and you can either install locally or start with a hosted Computer that runs in the cloud and opens in the browser. ### Which coding agents does omg.dev support? Settings, then Coding agents lists Claude Code, Codex, Grok, Cursor, omg agent, OpenCode, fx, Muse, DeepSeek, Devin, Jcode, Copilot and Pi. You sign in with the agent provider, and the harness uses your existing subscriptions or API keys rather than reselling access. ### Is the omg.dev local server authenticated? No. It has no built-in authentication and binds to 127.0.0.1, and the README says not to expose it to the public internet. Phone access goes through Tailscale with OMG_TAILSCALE_SERVE=1, and docs/remote-access.md plus SECURITY.md are the files to read before sharing access. ### How do I uninstall omg.dev from my own machine? Run omg computer uninstall, which removes omg.dev while keeping your sessions and config. The same group offers omg computer status to check the install, omg computer update for the latest release, and omg doctor for a sanitized diagnostic report. ### How do I run omg.dev from source? Clone the repository, run bun install, copy .env.example to .env, then bun run serve. The API is at http://localhost:8766, the UI needs cd web and bun run dev in another terminal, and bun run serve on its own requires a built web/dist. ## Sources - [Official documentation](https://omg.dev) - [Official README](https://github.com/BennyKok/omg.dev#readme) - [Project repository](https://github.com/BennyKok/omg.dev) - [Release notes](https://github.com/BennyKok/omg.dev/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/bennykok-omg-dev