# Connect your agent

> Nolan Edit comes with nolanlabs. Or run Claude Code, Codex, Gemini CLI or any ACP agent on its own sign-in: what each needs and how to add yours.

Page: https://nolanlabs.io/docs/agents

An agent comes with the app: Nolan Edit, on NolanX, the model nolanlabs runs. Or the chat runs one you already have, on your own account. This page says which agents work and what each needs.

## Nolan Edit

Nolan Edit is the agent built into nolanlabs, and NolanX is the model it runs on. The first time the app opens it offers it, and from then on it's in the picker at the top of the chat, with the nolanlabs mark and **NolanX**, its model, beside it. There's nothing else to install and no other account: the app sets it up by itself, and until that's done the picker says **Getting ready…**.

- **Credits.** NolanX's use is counted in credits. Every account has 100 a month, with no plan. A small change, such as a title or a trim, takes about 4. While it's picked, the chat's header shows what's left after the model's name, and **Settings** > **Account** shows this month's and your plan, with **See plans** (or **Manage plan** on a paid one). At zero it stops until the first of the next month, or until you move to a bigger plan, and says so; the editor and your other agents keep working. There's a daily limit too. Plans that hold more are on [Pricing](https://nolanlabs.io/pricing), and buying, changing and canceling one is in [Plans and billing](https://nolanlabs.io/docs/plans).
- **What it asks first.** It edits the timeline, reads files and looks at frames without asking. A shell command, a file it writes, the web, bringing media in, rendering and settings each need your answer in the chat: **Allow once**, **Always allow** or **Deny**.
- **What leaves your computer.** What you write, what it reads in the project and the frames it looks at go through nolanlabs.io to the company that runs NolanX. nolanlabs.io keeps counts of use, never what was in it, and sends those counts to PostHog, where nolanlabs reads them. See the [Privacy Policy](https://nolanlabs.io/privacy#nolanx).
- **In a studio.** A file named `policy.json` in the app's folder saying `{"models": "off"}` turns Nolan Edit off on that computer, and then it isn't offered.

Nolan Edit runs on Claude Code, which the app installs in your home folder. It needs no Claude account.

## Your own agent

Pick one at the top of the chat. The picker lists each agent and how it stands on this computer.

| Agent | What it needs |
| --- | --- |
| **Nolan Edit** | A nolanlabs account, signed in. See above |
| **Claude Code** | Claude Code installed and signed in on this computer |
| **Codex** | Codex installed and signed in (`codex login`), and its ACP adapter, `codex-acp` |
| **Gemini CLI** | Gemini CLI installed, with a paid Gemini API key or a Gemini Code Assist Standard or Enterprise plan. Google [ended personal sign-in](https://developers.googleblog.com/an-important-update-transitioning-gemini-cli-to-antigravity-cli/) (free, AI Pro and Ultra) on June 18, 2026 |
| Your own | Any agent that speaks ACP, the Agent Client Protocol, and its command |

There’s no key to paste into nolanlabs. Every agent gets the same eight tools, so each works through the same ops as the buttons. Question and plan cards are Nolan Edit’s and Claude Code’s.

## What its state means

The picker shows a note beside each agent.

| Note | What it means |
| --- | --- |
| **Signed in**, or your plan’s name | It’s installed and signed in. Pick it and ask |
| **Not signed in** | It’s here, but your account isn’t. Sign in with the agent’s own command |
| **Not installed** | It, or the piece that connects it, isn’t here, such as Codex without `codex-acp` |
| **Not on this computer** | Claude Code isn’t installed |
| **Sets itself up when you pick it** | Nolan Edit before its first use. Pick it and the app installs what it runs on |

## What comes along with Claude Code

The Claude Code agent is a full Claude Code session in the project’s folder, with the editor’s tools added. It isn’t capped.

- Its own shell, files, web, subagents and todos.
- Your own Claude Code settings, `CLAUDE.md` and skills (those in `~/.claude`), and the editor’s own skills.
- From the project’s folder, only its `CLAUDE.md` and `.claude/skills`. Its settings, hooks, MCP servers, `CLAUDE.local.md` and `.claude/rules` aren’t read, so a folder someone sent you can’t run anything by itself.
- No limit on turns and no tool timeout of its own. The conversation compacts itself as it grows.

## Add your own

Choose **Add your own agent…** at the bottom of the picker, or edit `agents.json` in the app’s data folder:

```json
{
  "default": "claude",
  "agents": [
    { "id": "my-agent", "name": "My agent", "command": "/usr/local/bin/my-agent", "args": ["--acp"] }
  ]
}
```

`default` is the agent the chat starts with. The editor gives every agent short instructions with the first message of each conversation, so one that knows nothing of the editor still works the same way.

## What the agent can reach

The timeline changes only through the proposal, which you can undo as one step. See [Proposals](https://nolanlabs.io/docs/proposals).

Bringing media in, rendering, changing settings and using its own shell, files and web all need your answer in the chat, unless they only look. **Project settings › Agent may** switches that to “Use everything without asking”. The agent can’t change that setting itself.

Each open project has its own agent processes, running in that project’s folder, and none can reach another’s project.

Frames the agent looks at go to the agent you chose, as do pictures you paste into the chat. Your footage itself stays where it is on your disk.

## Next

Make [your first cut](https://nolanlabs.io/docs/first-cut), or read about [the chat](https://nolanlabs.io/docs/chat).
