> For the complete documentation index, see [llms.txt](https://sync.wiki.mksdtech.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://sync.wiki.mksdtech.com/collaboration/ai-agents.md).

# AI Agents

Your own AI agents, each in its own Rhino, that take your requests in the Agents tab of the chat and model alongside you.

Ask for trees along a street, a column grid or three roof options, and keep modelling while it gets built. AI agents are helpers that run in Claude Code: each joins your session in its own Rhino, builds what you ask, and shares it like a teammate. They **Send** it, and you **Reload** to get it.

Agents are private. The agents you start work only for you: only you can ask them, and only you see their answers. Each person in a session can start their own. They appear in **Users** as **HAL-1**, **HAL-2** and so on, with an **AI** tag.

## What you need

* **Windows** and **Rhino 8.21 or later**. You can't start agents on macOS.
* **Claude Code**, connected to Rhino Sync. See [Set up](#set-up).
* **The Claude app** (your Claude plan) or **an Anthropic API key**.
* **A session with** [**quick changes**](https://sync.wiki.mksdtech.com/collaboration/quick-changes.md) **on.** If they're off, your agent asks before turning them on.
* **1-2 GB of memory** for each agent's Rhino.

## Set up

Click **AI agents**, the robot button next to the chat box, to open the **Agents** window. The first time, it shows two steps:

<figure><img src="https://2088031227-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMFZITZJW4MHLYhDLac3k%2Fuploads%2Fgit-blob-6159c302e2d17e4d78252130790b9e637d4fe4b0%2Fmock-agents-setup.png?alt=media" alt="The Agents window with Claude Code installed and Connect Claude still to do" width="412"><figcaption></figcaption></figure>

1. **Claude Code**: click **Get Claude Code** and install it.
2. **Rhino Sync agents**: click **Connect Claude**.

The Windows installer can do step 2 for you: tick **AI agents for Claude Code** on its components page. If the agents in Claude Code are older than your Rhino Sync, step 2 shows **Update**.

## Start your agents

Join a session first, then open the **Agents** window. It remembers the way you chose last time.

Your agents can run two ways. Pick the one that suits you:

|               | Claude app                                              | API key                                                           |
| ------------- | ------------------------------------------------------- | ----------------------------------------------------------------- |
| **Paid with** | Your Claude plan                                        | Your Anthropic API key, per use                                   |
| **Runs from** | A Claude Code chat you keep open                        | Rhino, in the background                                          |
| **You need**  | The [Claude app](https://claude.ai/download), signed in | A key from [console.anthropic.com](https://console.anthropic.com) |

### With the Claude app

<figure><img src="https://2088031227-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMFZITZJW4MHLYhDLac3k%2Fuploads%2Fgit-blob-a28d6269cebbb83dfff6f830beee6df7a356eef0%2Fmock-agents-start.png?alt=media" alt="The Agents window with Claude app chosen under START WITH, and Start" width="412"><figcaption></figcaption></figure>

1. Choose **Claude app** and click **Start**. A Claude Code chat opens with the request filled in.
2. Press Enter in Claude.
3. Go back to Rhino. Keep the Claude chat open: it runs your agents.

### With an API key

<figure><img src="https://2088031227-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMFZITZJW4MHLYhDLac3k%2Fuploads%2Fgit-blob-8b24f7a15b73aa24bb6ab79892a54d8c40e46919%2Fmock-agents-key.png?alt=media" alt="API key chosen, with the key saved and Opus 5.5 as the model" width="412"><figcaption></figcaption></figure>

1. Choose **API key** and paste your key in **Key**. Rhino Sync saves it on this computer.
2. Pick a **Model**: **Opus 5.5** or **Sonnet 5.5**.
3. Click **Start**. Nothing else to keep open.

Your agent joins the session and says hello in the **Agents** tab.

## Ask your agents

Once your agents are in the session, the chat has two tabs: **Team** for everyone, and **Agents** for you and your agents only. A bar beside the chat box shows where you're writing: green for **Team**, red for **Agents**.

<figure><img src="https://2088031227-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMFZITZJW4MHLYhDLac3k%2Fuploads%2Fgit-blob-dd9fee94b4e42809ae0d574dd6b97e22a444c8c3%2Fmock-chat-agents.png?alt=media" alt="The Agents tab: Mikael asks for a canopy, HAL-1 answers with a point and 4 steps · 2m 10s, and is working on the next request" width="412"><figcaption></figcaption></figure>

In **Agents**, write what you want, no `@` needed:

> `add trees along the east street, every 8 m`

* The agent marks your request **On it**, and replies under it when it's done.
* To say where, add a **Point**, your **View** or your **Selection** with **+**. See [Share a point, a view or a selection](https://sync.wiki.mksdtech.com/collaboration/sync-chat.md#share-a-point-a-view-or-a-selection).
* A request sent while your agent is busy waits its turn. A message about the work in hand changes it.
* For several things at once, write `add an agent`. You can have up to four.
* Write `sync` to have an agent save what's been sent to the central file. Agents never sync unless you ask.

## Watch and stop them

While your agents work, `working...` shows above the chat box. Click it to open the **Agents** window:

<figure><img src="https://2088031227-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMFZITZJW4MHLYhDLac3k%2Fuploads%2Fgit-blob-1887655f8d32416095b22a670f953a8160f26393%2Fmock-agents-working.png?alt=media" alt="HAL-1 working through its plan, HAL-2 idle" width="412"><figcaption></figcaption></figure>

Each card shows the agent's task, what it's doing now and its plan. Its buttons:

| Button                | What it does                                   |
| --------------------- | ---------------------------------------------- |
| **Show in the model** | Zooms to what the agent is working on.         |
| **Stop**              | Stops this request. You can also write `stop`. |
| **Close**             | Closes this agent and its Rhino.               |

Your agents leave when you do. When you end the session or close Rhino, agents started with the Claude app finish what you asked, then close. Agents started with an API key stop at once.

## Good to know

* **Their edits work like a teammate's.** Objects an agent is editing are in use until it sends, and it can't change objects you're editing. See [Edit conflicts](https://sync.wiki.mksdtech.com/collaboration/edit-conflicts.md).
* **Send before you ask.** Agents reload before each request, so they see only what you've sent.
* **Be specific.** Sizes, positions, layers: `12 columns, 400 mm square, on the grid at level 1`.
* **Leave their Rhinos alone.** Each agent's Rhino runs minimised, with **Autopilot** in its **Sync** panel.
* **What Claude sees:** what the agents report about the model, your requests and the views they capture, through your Claude app or API key. The Rhino Sync server never sees your model.
