odette guide
odette runs Claude Code agents in sandboxes on your machine, and checks anything they try to reach or run that you have not allowed: an AI judge decides what it can, and asks you about the rest. This guide covers installing it, running agents, and deciding what they may do. Start at the top if you are new.
| Page | What it covers |
|---|---|
| Getting started | Install odette, set it up, start your first agent, approve its first request and check that the sandbox holds |
| How it works | Understand the parts, follow one request through them, and see when odette fits compared with other ways to run agents |
| Workflows | Follow a task from start to finish: leave an agent working, run several at once, or roll odette out to a team |
| Sessions | Run agents in parallel, continue a conversation, and manage Claude homes, extra folders, profiles, service API keys and Google Workspace |
| Approvals | Choose how much odette checks an agent, allow or deny hosts and commands, and answer agents’ questions |
| Dashboard | Find your way around the web dashboard, its notifications, sign-in and the terminal status row |
| Pods | Let sessions message each other, with a supervisor directing the rest |
| Configuration | Change global settings and per-project files, add plugins, and switch feature flags |
| Kits | Set a project up in one command with a reusable kit, and keep your own kits in a repository |
| Security | See what odette encrypts, what an agent can and cannot reach, and where its protection stops |
| Upgrading | Update odette to a new release, or uninstall it and remove everything it stored |
| Troubleshooting | Find the fix for a problem by its symptom, from setup errors to blocked agents |
| Commands | Look up every odette command and flag |
Words used throughout
Section titled “Words used throughout”The dashboard, the commands and this guide use these words for the same things.
| Word | Means |
|---|---|
| Session | One Claude Code agent running in its own container. |
| Pod | Sessions started with the same --pod name, which can message each other. |
| Profile | One Claude sign-in, with its service API keys and its Claude homes. --profile picks one (the only signed-in one is used without it); odette profiles lists them. Not the same as a Google Workspace account. |
| Claude home | The folder on your machine holding a profile’s Claude data for one project: conversations, memory, settings. |
| Gate | A check on what an agent does: the Commands gate for commands it runs on your machine, the Egress gate for every network connection it makes. |
| Gate mode | How a gate decides: manual, judge, yolo+judge or yolo (Approvals). |
| AI judge | The model that decides requests in the judge and yolo+judge modes. |
| Inspected host | A host whose traffic odette decrypts so you can approve it by path (Approvals). |
| Dashboard | The web page where you see sessions and answer what is waiting on you. odette dashboard opens it. |
| Daemon | odette’s background service: the proxy, the gates and the dashboard. A launch starts it. |
| Data folder | ~/.odette/, where odette keeps its settings and stored secrets. |