20. Settings — every tab of the ⚙ settings dialog
How each setting is used belongs to the other chapters, so read this one as a map.
Opening it, and how it is organised
Section titled “Opening it, and how it is organised”⚙ Settings is in the account menu of the top bar. The left rail is split into four groups.
| Group | What is in it |
|---|---|
| Personal | Display / Account / Keyboard / Read aloud / Notifications / Limit board / Assistant / AI assistance / Agent instructions / Agent memory / Memory snapshots |
| Connections | Agents / Git hosting / Ops & monitoring / Issue tracker / Chat / MCP servers (used by sessions) / AWS profiles/SSM / Google Cloud |
| Outside access | MCP tokens & apps (personal access tokens and connected apps that drive this Workspace from outside) |
| Workspace | Agent usage / Cloud cost / Running time / Machine / Toolchains / Databases / Preview subdomains / Internal repos / Export / import / Danger zone |
- It remembers the tab you opened last and reopens there.
- On a phone it is a list → detail drill-down. Back returns to the list; back again closes the dialog.
- Managing your team’s members, quotas and audit is a separate dialog, Tenant settings in the same menu, shown only to those with the permission (admin/). Admin (the shield icon) is the deployment-wide one.
- Some tabs need the workspace running, because the setting itself lives inside the workspace. While it is stopped you get a note and a “Start workspace” button.
When a change takes effect
Section titled “When a change takes effect”Getting this wrong is what makes a setting look like it “didn’t work”.
| Timing | What |
|---|---|
| Immediately | Display, keys, speech, notifications; adding and removing connections; the stopped-session archive period (on the next session-list refresh) |
| From the next session you start | Agent behaviour settings, agent instructions, session-to-session messaging, starting sessions from sessions, past-session search (turning it off also refuses running sessions’ searches at once), Agent Fleet memory (likewise), image generation, MCP servers |
| From the next chat message | Assistant settings; ops & monitoring connections (when used from an assistant) |
| After stopping and starting the workspace | Toolchains (timezone, language versions); Machine (a size or class your admin changed) |
There are also two storage scopes. The theme, the surface colours and the main-area layout are stored on this device only; everything else (font, font size, …) is stored on the server and follows you to another PC or browser. If saving to the server fails, the top of the settings screen says your settings could not be saved; until then the agents act on the previous settings, so press Retry.
Personal
Section titled “Personal”Display
Section titled “Display”Language, time zone, theme, fonts and the main-area layout. See Change the display and your account.
Account
Section titled “Account”Adding and removing sign-in methods. See Change the display and your account.
Keyboard
Section titled “Keyboard”Shortcut assignments, terminal input priority, the send key and reply suggestions are in Use the keyboard and the command palette.
Read aloud
Section titled “Read aloud”Reads answers out loud. See Use read-aloud.
Notifications
Section titled “Notifications”Choose the unread dot, the OS notification and read-aloud for each kind of notification. See Get notified.
Limit board
Section titled “Limit board”Choose who in this tenant sees where you stand on usage limits, and how much. See Show each other where you stand on usage limits.
Assistant
Section titled “Assistant”Reply language, agent priority, the assistant models, auto-compaction and so on. See Ask an assistant. Auto-replies to session reports and autopilot are in Hand work to the Fleet Operator.
AI assistance
Section titled “AI assistance”The agents and models used for one-shot AI generation such as titles and reply suggestions, and turning each feature on or off. See Use translation and purpose-built assistants.
Agent instructions
Section titled “Agent instructions”Adds your own standing instructions to every agent started afterwards. See Write down how you work, once.
Skills
Section titled “Skills”Lists each agent’s personal skills, read-only. See Bring your Claude Code setup.
Agent memory
Section titled “Agent memory”Agent Fleet’s own shared memory, which every kind of agent reads and writes, with the sessions’ switch in its Settings view. The page is described in Read and correct agent memory.
Memory snapshots
Section titled “Memory snapshots”Version control over the memory an agent keeps in its own files. The page is described in Restore memory from a snapshot.
Connections
Section titled “Connections”Agents
Section titled “Agents”Each agent’s card connects it and sets its behavior. See Choose an agent and each agent’s page.
The
Sessions group holds when stopped sessions are archived (Default, which is the deployment’s period (7 days unless the deployment changed it), 1 / 3 / 7 / 14 / 30 days, or Off), session-to-session messaging, starting sessions from sessions (with
children per session), past-session search (on by default: lets a
session search this workspace’s past conversations with the search_sessions tool; the command palette’s
“Conversations” mode works either way, see 02), image generation and the image provider order (this deployment’s own engines first, each under its own name, then the CLI routes).
→ Let sessions message each other,
Start child sessions from a session,
Have a session make images
Git hosting
Section titled “Git hosting”The GitHub and Bitbucket connections, the commit author, and the saved Subversion credentials. Connecting GitHub also connects GitHub Copilot. See Connect a Git service.
Ops & monitoring
Section titled “Ops & monitoring”Connects PagerDuty, Grafana, CloudWatch and AWS to the SRE assistant. How to connect them, and what the assistant does with them, is in Connect monitoring tools.
Issue tracker
Section titled “Issue tracker”The Jira connection behind the left rail’s Issue tracker. Connecting it and saving queries are in Connect an issue tracker.
Connects a Discord or Slack bot and sets which notifications it sends. How to connect it, and how to answer by replying, is in Connect Discord and Slack.
MCP servers (used by sessions)
Section titled “MCP servers (used by sessions)”The MCP servers your sessions and assistants use: registering one, the three origins, the connection test and the OAuth sign-in are in Add an MCP server to your sessions. To let an outside app drive this Workspace, see MCP tokens & apps under Outside access.
AWS profiles/SSM
Section titled “AWS profiles/SSM”Your AWS profiles (shared settings), which af-aws-exec and SSM sessions sign in with, and SSM hosts (individual)
for logging in to another in-house host. A profile is IAM Identity Center (SSO) or Assume a role from another
profile; an SSO row has Log in and, when signed in, Log out; an assume-role row’s Log out ends its source’s login, after a confirmation that names it. How logging
in and out work, and the two types, is in Use AWS SSO; SSM hosts are in
Log in to EC2 with SSM.
Google Cloud
Section titled “Google Cloud”Your Google Cloud profiles, which af-gcloud-exec runs commands with. Adding, importing and logging in, and the WS
bar’s Google Cloud badge, are in Use Google Cloud.
Outside access
Section titled “Outside access”MCP tokens & apps
Section titled “MCP tokens & apps”Personal access tokens and connected apps for using this workspace from outside. Tokens are in Drive your workspace from an external Claude, connected apps in Connect a remote MCP client by signing in. To register a server your sessions call, see Add an MCP server to your sessions.
Workspace
Section titled “Workspace”Agent usage
Section titled “Agent usage”The ledger of what your tokens went on. How to read it is in Check prices and usage.
Cloud cost
Section titled “Cloud cost”The cloud cost attached directly to your workspace (deployments on AWS only). How to read it is in See running time and cloud cost.
Running time
Section titled “Running time”When your workspace was running, as a grid of hours. How to read it is in See running time and cloud cost.
Machine
Section titled “Machine”The size of the machine your workspace runs on, its memory and disk usage, and how to free disk space (including deleting your old ~/.af-work folders by age):
See your machine and its usage.
Toolchains
Section titled “Toolchains”The time zone, the Node.js / Java / Go versions, tool versions and agent CLI updates: Choose your toolchains.
Databases
Section titled “Databases”This workspace’s PostgreSQL and MySQL: Run database-backed tests.
Preview subdomains
Section titled “Preview subdomains”The tab for opening the app you are building at its own URL. It is only there on a deployment that issues preview subdomains. The ports, sharing and re-issuing the URL are in Use preview subdomains; the development OIDC IdP block (issuer, client id, app configuration examples, test personas) is in Try Google sign-in on a preview.
Internal repos
Section titled “Internal repos”Create Git repositories inside the tenant, with no external service. See Use the built-in Git.
Export / import
Section titled “Export / import”Your settings in one file, to take to another environment and read back: Export and import your settings.
Danger zone
Section titled “Danger zone”Recreating the workspace and cleaning home (on deployments that can). What stays and what goes is in Recreate the workspace.
What you want to do → which tab
Section titled “What you want to do → which tab”| What you want | Tab |
|---|---|
| Text is too small / change the colours | Display |
| Sign in with another method too | Account |
| Send with Enter | Keyboard (send key) |
| Switch panes by tab | Display (main area layout) |
| Rebind a shortcut / pass keys to the terminal | Keyboard |
| Have replies read out loud | Read aloud |
| Silence Slack without disconnecting | Notifications (service notifications) |
| Make the assistant answer in English | Assistant (output language) |
| Let the operator run ahead, or stop it | Assistant (autopilot, auto-reply) |
| Stop AI from writing titles or reply suggestions | AI assistance (per-feature on/off) |
| Run only the assistance on a cheap model | AI assistance (priority, models) |
| I keep typing the same preamble | Agent instructions |
| Sign in to Claude / Codex | Agents |
| Stop a model that bills extra from being picked | Agents (models to exclude) |
| Keep stopped sessions in the list longer, or out of it sooner | Agents (archive stopped sessions after) |
| Let sessions talk to each other | Agents (session-to-session messaging) |
| Stop sessions from searching past conversations | Agents (past-session search) |
| Let agents share what they learn across kinds | Agent memory (Settings) |
| Let a session see what the other sessions are doing | Nothing to set: every session can (Let a session see the fleet) |
| Have a session leave you a note about what it noticed | Nothing to set: every session can (Let a session see the fleet) |
| Clone a private repository | Git hosting |
| Get help investigating an incident | Ops & monitoring |
| Start work from a ticket or an issue | Issue tracker |
| Follow progress while away from the desk | Chat |
| Let the AI use an in-house tool | MCP servers (used by sessions) |
| Drive it from the Claude on my laptop | Outside access → MCP tokens & apps |
| Find out what the tokens went on | Agent usage |
| Find out what my workspace costs to run | Cloud cost |
| I want to know when my workspace was actually running | Running time |
| It memorised something wrong | Agent memory (Memories, Changes); Memory snapshots (History) |
| Change the Java / Node version | Toolchains |
| Run a test suite that needs a database | Databases |
| Show the app you are building to someone | Preview subdomains |
| Get into another server | AWS profiles/SSM |
| Run commands in my Google Cloud projects | Google Cloud |
| Keep code that cannot leave the building | Internal repos |
| Take my settings to another environment | Export / import |
| The environment is broken | Danger zone |