Find what you want to do
Each page covers one thing you want to do. The chapters of the current guide are being rewritten into these pages. Anything not here yet is in the chapters of Using Agent Fleet.
Getting started
Section titled “Getting started”| Page | What you can do |
|---|---|
| What you can do with Agent Fleet | Learn what it does and which page to read first |
| Start your first session | Go from signing in to the first reply |
| Terms and the big picture | See how tenants, workspaces, sessions and agents relate |
| Find your way around the screen | Know what sits where on the top bar, the workspace action bar, the left pane and the main area |
| Use the keyboard and the command palette | Drive the Console without the mouse, with shortcuts and the command palette |
| Split the screen into panes | Lay out sessions and files in panes, and choose the layout |
| Use it on a phone | Check progress and reply from a phone |
| Stop the workspace at the end of the day | Stop the workspace with your work kept, and deal with Restart needed |
| Start here if you don’t use the terminal | Use only chat, reading and progress checks |
Sessions
Section titled “Sessions”| Page | What you can do |
|---|---|
| Start a session | Launch a session from Start, choosing the agent, the model and a first prompt |
| Work in parallel with worktrees | Give each session its own working copy and work in parallel in one repository |
| Start from an issue or a ticket | Start a session from an issue or a pull request in the issue tracker list |
| Read a session’s state | Tell from the row marks and notifications which sessions are waiting for you |
| Reply to a session | Read the conversation, answer question, plan and permission cards, and send the next instruction |
| Hold what you send, stop a turn | Hold follow-ups during a turn and send them in order, and stop a turn when something is queued |
| Comment on a plan, have it reviewed | Send comments on parts of a plan, and have another agent review it |
| Call a skill or a command | Pick a skill or a command the session can call and put it in the input box |
| Branch from a past message | Start a new session holding the conversation up to one of your past messages |
| See what a session changed | See the files a session actually touched, as a list |
| Save a conversation as Markdown | Download the whole conversation as a Markdown file |
| Read an answer in your own language | Read an answer in a language you do not read, translated without using a turn |
| Stop and resume a session | Pause a session and later carry on from where it stopped |
| Stop when done, or keep a session awake | Have a session stop once its work is done, or keep a long job from being stopped |
| Clear away finished sessions | Archive sessions and delete working copies and branches you no longer need, and restore them from the trash |
| Stop a session that spends too much with a budget | Stop a session when its estimated spend reaches a budget |
| Hand off to another session or agent | Continue a conversation in a new session or with another kind of agent |
| Page | What you can do |
|---|---|
| Connect a Git service | Connect GitHub or Bitbucket to clone and push private repositories |
| Clone a repository | Bring a repository into the workspace, or start from nothing |
| Use the built-in Git | Keep repositories inside the tenant, with no external service |
| Launch a session from a repository | Launch a session from a repository row and read the row’s badges |
| Initialize Git Flow | Make the branches started from issues follow git-flow |
| Commit in the commit graph | Read history and diffs, and commit without opening a terminal |
| Use a Subversion repository | Check out a Subversion repository and work in it from a session |
Files and terminal
Section titled “Files and terminal”| Page | What you can do |
|---|---|
| See and download files | Walk the workspace’s files, open them, and download files and folders |
| View documents and diagrams | Read Markdown, slides, .drawio diagrams, PDFs and Office documents in the Console |
| Edit a file | Fix an open file in place and save it. Have an AI propose a change |
| Hand a file to a session | Send a selection, with instructions, to a session or an assistant |
| Use the terminal | Copy and paste in the browser terminal, read its lag, change its font |
Preview
Section titled “Preview”| Page | What you can do |
|---|---|
| See a web app you started | Open a web app started in the workspace in the browser pane or the lightweight preview |
| Use preview subdomains | Open the app at its own root URL and show it to colleagues or outsiders |
| Try Google sign-in on a preview | Try Google sign-in on a preview URL or from an app on your own machine |
| Operate a browser the agent opened | Take over a step only a person can do, such as a login, in the agent’s browser |
Agents
Section titled “Agents”| Page | What you can do |
|---|---|
| Choose an agent | Pick and connect the agents that match your subscriptions, and set default and excluded models |
| Connect Claude | Run Claude Code in sessions, and use Remote control |
| Connect Codex | Run Codex on a ChatGPT subscription or an API key |
| Connect OpenCode | Switch between models from many providers with OpenCode |
| Connect GitHub Copilot | Run the Copilot CLI on your GitHub connection alone |
| Connect Cursor | Run the Cursor CLI on your Cursor account |
| Connect Kiro | Install Kiro and run it on a Builder ID or similar |
| Connect Antigravity | Run the Antigravity CLI on a Google account (experimental) |
| Connect Muse Code | Install Muse Code and run it on a Meta account or an API key |
| Use lcpp | Run sessions on the deployment’s llama.cpp engine or your own llama-server |
| Check prices and usage | See model list prices, what is left of your allowance and what your tokens went on; save with RTK |
| Write down how you work, once | Write your standing requests once so that every agent gets them |
| Bring your Claude Code setup | Use your own skills, subagents and hooks in the workspace |
| Page | What you can do |
|---|---|
| See every running session | Lay out the running sessions as cards and spot the ones waiting for you |
| See how sessions relate over time | See on a time axis which session started which, and what passed between them |
| Narrow the view with working sets | Group repositories and conversations per piece of work and switch what the left pane shows |
| Start child sessions from a session | Have a session start other sessions and hand them parts of the work |
| Let sessions message each other | Have sessions running side by side send each other short messages |
| Tell a session about its pull request’s review | Tell the session that opened a PR about failed CI and review comments |
| Let a session see the fleet | Let a session see the other sessions’ state and leave you notes |
| Hand work to the Fleet Operator | Have it start sessions, instruct them and take their reports, from chat |
| Ways to use the Fleet Operator | Patterns that work, such as parallel implementation and cross-review |
| Run at a set time | Have an agent work at set times or intervals |
Integrations
Section titled “Integrations”| Page | What you can do |
|---|---|
| Use AWS SSO | Have agents run AWS commands with your permissions |
| Log in to EC2 with SSM | Log in to an EC2 host from a terminal in your browser |
| Use Google Cloud | Have agents run Google Cloud commands with your permissions |
| Add an MCP server to your sessions | Let agents use an in-house tool or an outside service as an MCP server |
| Connect an issue tracker | List your Jira issues and GitHub issues in the left rail and start sessions from them |
| Use Jira | Have agents read, file and update Jira issues |
| Use Confluence | Have agents find and write Confluence pages |
| Use the GitHub CLI | Have agents create and read GitHub pull requests with gh |
| Use the Bitbucket CLI | Have agents create and read Bitbucket pull requests and check pipelines with bkt |
| Connect Discord and Slack | Get session notifications in Discord or Slack, and answer by replying or with buttons |
| Connect monitoring tools | Have the assistant help sort out an incident while it looks at PagerDuty, Grafana, CloudWatch and AWS |
Images
Section titled “Images”| Page | What you can do |
|---|---|
| Use the image gallery | Look through a folder’s pictures as cards |
| Generate images | Make pictures in bulk on the deployment’s engine, without an agent |
| Finish an image in the studio | Refine an image-generation draft by talking to an agent |
| Have a session make images | Have an agent make a picture during a conversation |
Workspace
Section titled “Workspace”| Page | What you can do |
|---|---|
| Recreate the workspace | Rebuild a broken environment from the latest image |
| See your machine and its usage | Check the machine’s size and how memory and disk are used, and free disk space |
| Choose your toolchains | Pick the Node.js, Java and Go versions and the time zone |
| Run database-backed tests | Run tests against PostgreSQL and MySQL inside the workspace |
| Try AWS code against a local emulator | Try AWS code without touching a real account |
| Export and import your settings | Take your settings to another environment or account |
| See running time and cloud cost | See when the workspace was running and what it costs |
| Page | What you can do |
|---|---|
| Hand a session to a member | Have a member you share a session with carry it on |
| Ask a teammate through the inbox | Send members a message or a request; whoever takes it starts a session |
| Share a conversation | Show a conversation read-only, and take suggestions on it if you want |
| Show each other where you stand on usage limits | Show members you choose how close you are to your usage limits |
| Page | What you can do |
|---|---|
| Ask an assistant | Ask questions without a repository; compact a long conversation and keep a work plan |
| Use translation and purpose-built assistants | Translate or summarize a file; make your own assistant and set up AI assistance |
| Collect memos and hand them over | Collect instructions as they come, tidy them, and send them to a session together |
Memory
Section titled “Memory”| Page | What you can do |
|---|---|
| Choose which memory your agents use | Tell instruction files, the agent’s own memory and Agent Fleet memory apart, and choose which to use |
| Read and correct agent memory | Read, rewrite, pin and forget what agents remember |
| Check memory changes | See who changed what and when, and revert a bad change |
| Set how memory is used | Let sessions use memory, and carry Claude Code’s memory across |
| Find stale memories | Check for memories that use missing paths or retired terms, and mark them |
| Use memory tidy-up | Have a model draft fixes to memory, and approve them one by one |
| Import and export memory | Bring Claude Code’s or muse’s memory in and write it back to Claude Code; move the memories of a project that no longer exists |
| Restore memory from a snapshot | See the history of Claude Code’s and Codex’s own memory, restore a point in time, take it to another environment |
| Use memory from an external agent | Use Agent Fleet memory from your own Claude Code or claude.ai; bring your own memory in |
Outside access
Section titled “Outside access”| Page | What you can do |
|---|---|
| Drive your workspace from an external Claude | Drive sessions from your own Claude Code or Claude Desktop with a token |
| Connect a remote MCP client by signing in | Connect claude.ai, ChatGPT, Claude Code or Codex without a token |
| Move work from your own Claude Code into the workspace | Push work you started on your computer and carry on in a workspace session |
Settings
Section titled “Settings”| Page | What you can do |
|---|---|
| Change the display and your account | Change language, time zone, theme and fonts; add a sign-in method |
| Get notified | Choose the unread dot, the OS notification and read-aloud for each kind of notification |
| Use read-aloud | Have answers read out loud; choose the voice and how it reads |
Security
Section titled “Security”| Page | What you can do |
|---|---|
| Handle credentials safely | Know where tokens and keys belong, and where never to write them |
Troubleshooting
Section titled “Troubleshooting”| Page | What you can do |
|---|---|
| Fix a workspace that will not start or has stopped | Find the fix from the badge beside the power button, or a start or disk message |
| Find a missing session or one that will not resume | Fix sessions that vanished, will not resume or stopped partway, and sharing or scheduled runs that misbehave |
| Fix an agent’s login or a missing model | Fix an expired login, a model you cannot pick, or an assistant that stopped |
| Fix a failed clone or Git authentication | Fix failed clones, private-repository authentication and submodule fetches |
| Fix a preview or browser pane that does not show | Find the fix from the error the browser pane or a preview URL shows |
| Fix MCP, cloud and Jira integrations | Fix MCP servers, AWS and Google Cloud commands, jira, and apps connected from outside |
| Fix the terminal, keys and phone layout | Fix keys, copy and paste, the phone layout, and menus that do not appear |
| Read the frequently asked questions | Read what to try first, and answers to common questions |
| Report a bug or request a feature | Tell the project about a bug in Agent Fleet itself, or a feature you want, with an issue |