> For the complete documentation index, see [llms.txt](https://planetz.gitbook.io/planetz/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://planetz.gitbook.io/planetz/core-ideas/multi-agent.md).

# Multi-agent fleet

> Status: Core ideas · Target version 0.1.x

## What you'll learn

Why Planetz is built around a *squad* of agents instead of one chat thread, and how running work in parallel lanes changes your job from typing to directing.

***

## From one chat window to a deck

A single chat assistant does one thing at a time, in one conversation. Planetz runs a **fleet**: multiple autonomous agents, each working its own task in its own lane, all visible on one deck.

> Your role shifts from *writing every line* to **directing the squad** — reviewing, steering, and approving — while the work happens in parallel.

The deck shows the fleet at a glance: how many agents are active, and what each one is doing right now.

## Agents have roles

Work is divided across specialized roles rather than one generalist doing everything. Typical roles include:

* **Planner** — turns a task into a plan.
* **Coder** — makes the changes (in an isolated lane / branch).
* **Reviewer** — checks the work against intent and quality bars.
* **Tester** — runs and reproduces tests.

This **maker–checker** split is deliberate: the agent that writes the code is not the only agent that judges it. Separation is what keeps an autonomous run honest.

## Lanes and parallelism

Each running task gets its own isolated lane — its own branch and workspace — so agents don't step on each other. You can run many at once and let them progress independently. Each agent reports a live status:

| Status        | Meaning                                 |
| ------------- | --------------------------------------- |
| **idle**      | Available, no active task               |
| **working**   | Actively running a task                 |
| **reviewing** | Checking work (maker–checker)           |
| **waiting**   | Blocked — often waiting on a human gate |
| **error**     | Run hit a problem and needs attention   |

That single state vocabulary is the same language the optional **Manta** desktop robot speaks, so the fleet's state can be felt at a glance — on screen or on the device. See [Edge AI & data sovereignty](/planetz/core-ideas/edge-ai.md) and [How Planetz is different](/planetz/why-planetz/how-planetz-is-different.md).

## Bring your own agents

The fleet isn't limited to Planetz's built-in runtime. External agents — for example **Cursor** — can join the deck and push their status in, so a mixed squad of built-in and external agents is visible and governed from the same place.

## Why a fleet needs the ledger

Parallelism multiplies a danger: several agents, in several lanes, each quietly making decisions. Without a shared record, those decisions never get reconciled and the product drifts. That's why the fleet only delivers its promise when it's paired with the [Intent Ledger](/planetz/core-ideas/intent-ledger.md) — every agent's decisions land in one place, traced against one intent, where you can see conflicts and lock the calls you agree with.

## Where you see it in the product

* [**Task deck**](/planetz/the-product-screen-by-screen/task-deck.md) — queue work and watch the fleet.
* [**Execution logs & summary**](/planetz/the-product-screen-by-screen/logs-and-summary.md) — follow any single agent's run in detail.

## Next

* [The harness: govern, don't pre-approve](/planetz/core-ideas/harness-governance.md) — the workflows the fleet runs inside.
* [Git integration & disposable workspaces](/planetz/core-ideas/git-integration.md) — how lanes stay isolated and recoverable.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://planetz.gitbook.io/planetz/core-ideas/multi-agent.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.