Bring your own agent to Omne

How it works

A small connector program runs on your computer. Omne generates a command for it in Settings, under AI providers, and you run that command once. The connector links your computer to your workspace, then waits for your messages. When you pick the agent in Ask AI and send a message, Omne passes it to the connector, which hands it to the agent's program. The agent answers, and the connector sends the answer back to your chat.

While the agent is working it can use Omne's tools: it can search and read your Library, save to it with your approval, see which accounts you have connected, and use their tools. Those tools exist only for the length of a turn that you started, which is part of how Omne stays within the makers' terms. See What an agent can do in Omne.

What you need

• A computer that is switched on and online while you use the agent. The connector has a background service for Mac, Linux and Windows, so the agent stays online after you close the window. Only the Mac one has been run end to end so far.
• Node.js 18 or newer. The setup prompt tells the agent to install Node's current long-term release if it is missing.
• The agent's program installed and signed in on that computer. Each agent's panel lists the install and sign-in commands.

The short version of setting it up

1. In Settings, open AI providers and find Bring your own agent.
2. Open the agent's card and give it a name.
3. Copy the prompt Omne gives you and paste it into the agent on your computer. Approve the one command it asks to run.
4. Wait for Waiting for your agent to connect… to change to a message that the agent is connected.
5. Open Ask AI, click the model name, and choose the agent under Your agents.

The full walk-through is in Connect an agent.

Which agents

Omne lists Claude Code, Codex, Gemini CLI, Antigravity CLI, Grok Build, Hermes, GitHub Copilot CLI, Cursor CLI, OpenCode, goose, Qwen Code, Cline CLI, Kimi CLI, Factory Droid, Kiro CLI and Devin CLI, plus any other agent that speaks the Agent Client Protocol. The list and what is proven are in Supported agents.

What stays private

Omne keeps the agent's name and a key for the connection. It never sees or stores your sign-in with the maker. Only you can use your agent: a personal plan is for one person, so teammates in your workspace cannot reach it. See Billing, terms and who can use your agent.

What it can touch on your computer

By default, Claude Code and Codex are held to Omne's tools and an empty folder. They cannot read your files or run commands on your computer unless you change that with a command on that computer. Other agents follow their own settings, so Omne asks you to tick a box that says so before you connect them. Read What an agent can touch on your computer before you connect one.

Provider or agent

Use a provider when you want Omne to run the model through an API key and keep working with your computer off. Use an agent when you already pay for a plan with a maker, want its own tools and behavior, and are happy to keep the computer on. You can have both, and switch between them in the model menu. Providers are covered in What Ask AI is.