---
title: "Overlane Agent"
description: "The assistant in the app and the docs. It answers from the docs with sources, takes you to the right page, reads your workspace and proposes changes that you approve first."
---

> Documentation Index
> Fetch the complete documentation index at: https://docs.overlane.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Overlane Agent

Overlane does not run coding agents of its own. Your agents (Claude Code, Codex, Cursor, Gemini CLI and others) connect to it and it monitors their work. The only thing Overlane runs itself is an inference helper, like the Overlane Agent below and the decision model Clef, and only for a decision that rules, search and other deterministic tooling cannot make. Everything else is deterministic.

The Overlane Agent is not a coding agent and never works on its own. It uses a model only where plain code cannot answer, and it changes nothing without your approval. It is a chat panel in the app and on this site. Press `⌘J` (`Ctrl+J` on Windows and Linux) to open it, or use the **Ask the docs** button on any docs page.

It does four things, in this order of care:

1. **Answers questions from the docs**, quoting the pages and linking to them. If the docs do not cover a topic, it says so instead of guessing.
2. **Takes you to a page** in the app, from the list of pages your role and plan can open.
3. **Reads your workspace** (projects, issues, runs, reviews, costs, logs, diagnostics) with your own permissions, never more.
4. **Proposes changes** such as creating or assigning an issue, and waits. Nothing changes until you approve it.

Everything works on Overlane itself. A connection such as GitHub or Linear is optional: say "via github" in a request only if you use one.

- [Using the assistant](/overlane-agent/using-the-agent/): where to open it, what to type and which requests it understands.
- [Tools and permissions](/overlane-agent/tools-and-permissions/): every tool, what role it needs and what is never available.
- [Approvals and undo](/overlane-agent/approvals-and-undo/): the approval card, who can approve, expiry and undo.
- [Read-only mode](/overlane-agent/read-only-mode/): how an admin stops every change.
- [Citations and navigation](/overlane-agent/citations-and-navigation/): how answers are sourced and why a link is always real.
- [Privacy and retention](/overlane-agent/privacy-and-retention/): what is kept, for how long, and what the assistant never sees.
- [Planner model](/overlane-agent/planner-model/): the optional model for requests no rule understands, and what the owner registers for it.

The assistant is deterministic first: rules and search handle what they can, and a model is used only for a request that no rule understands, when one is configured. See [Teams and roles](/teams-and-roles/) for the roles it respects.

Source: https://docs.overlane.app/overlane-agent/index.mdx