Everything adf accepts, and every command and key of the terminal app, generated from the code. For task-oriented guides see the ADF CLI docs.
Command line
Usage: adf [--url <daemon-url>] [--json] <command>
adf [tui] [--view <id>] [--agent <id>] [--loop <name>]
adf daemon [start|status|stop|restart|logs|token]
With no command, adf opens the terminal app. Commands that talk to the daemon start a local one in the background first when none answers (not with --no-daemon / ADF_NO_AUTOSTART=1, a remote --url, or next to a running ADF Studio). <agent> is an agent id, handle or name.
Commands
| Command | Description |
|---|
adf tui | The terminal app (the default) |
adf agents | List loaded agents |
adf status <agent> | Show runtime status |
adf start <agent> | Start an agent and fire startup when applicable |
adf stop <agent> | Stop and unload an agent |
adf unload <agent> | Alias for stop |
adf loops <agent> | List the agent’s loops: main plus its inner (side) loops, parallel threads with their own history, e.g. a memory consolidator on a timer |
adf interrupt <agent> [--loop <name>] | End the running turn; the loop goes idle and keeps accepting chats, timers and triggers |
adf abort <agent> [--loop <name>] | Hard-abort the current turn without unloading; that loop stays stopped until the agent is reloaded (prefer interrupt) |
adf runtime [agent] | Show daemon or agent runtime diagnostics |
adf providers | Show provider configuration and agent resolution |
adf auth | Show auth and credential presence |
adf auth login <chatgpt|grok|claude> | Sign in to a subscription provider [—relay|—loopback] chooses where the ChatGPT OAuth callback is served; defaults to relay when —url points at a remote daemon. claude signs in through Claude Code on the daemon host (ADF stores no Claude credentials) |
adf auth logout <chatgpt|grok|claude> | Clear a subscription provider’s tokens (claude: signs Claude Code out on that host) |
adf settings | Show sanitized daemon runtime settings |
adf network | Show mesh and WebSocket diagnostics |
adf network mesh [enable|disable] | Control daemon mesh registration |
adf network server [start|stop|restart] | Control the mesh HTTP server |
adf network tools|lan|runtimes | Show mesh tools, LAN addresses, or peers |
adf usage [agent] | Show runtime or agent token usage |
adf config <agent> | Show agent config |
adf files <agent> | List agent files |
adf file <agent> <path> | Print one agent file |
adf inbox <agent> | List inbox messages |
adf outbox <agent> | List outbox messages |
adf timers <agent> | List timers |
adf tasks <agent> | List tasks and pending approvals |
adf task <agent> <taskId> | Show one task |
adf approve <agent> <taskId> | Approve a pending task |
adf deny <agent> <taskId> [reason] | Deny a pending task |
adf asks <agent> | List pending ask requests |
adf answer <agent> <requestId> [--loop <name>] <answer> | Answer a pending one-question ask |
adf answer <agent> <requestId> [--loop <name>] --answers '<json>' | Answer an ask form: a JSON array with one entry per question, in order: text or an option label, an array of labels for a multi-pick question, or null to skip |
adf identities <agent> | List identity metadata without secret values |
adf mcp [agent] | Show daemon MCP registrations or agent MCP state |
adf adapters [agent] | Show daemon adapter registrations or agent adapter state |
adf events [agent] | Follow daemon SSE events |
adf chat <agent> [--loop <name>] [--attach <file>]... <message> | Send chat (to main, or an inner loop) and print the accepted turn id. —attach uploads a file to the agent’s loop-upload/ folder and attaches it |
adf identity | Show the owner identity status |
adf identity new | Create an owner identity; shows the 12-word seed phrase once — write it down |
adf identity restore | Restore the owner identity from its seed phrase (typed at a hidden prompt; the same phrase as ADF Studio = the same owner) |
adf identity import [--i-have-a-backup] | Import an owner phrase like Studio’s Import identity: another owner’s phrase replaces this machine’s owner (asks first; needs the old phrase backed up) and restamps local agents |
adf identity unlock|lock | Unlock/lock a passphrase-protected identity (machines without an OS keychain) |
adf identity trust-daemon [--yes] | No OS keychain: let ADF Studio wrap agent credentials to this daemon’s key file, so they unlock without the passphrase (asks) |
adf identity untrust-daemon | Revoke that trust |
adf templates | List agent templates |
adf new [name] [--template <id>] [--provider <id>] [--model <id>] [--dir <path>] [--start] | Create an agent from a template, sealed with the owner identity (like Studio) |
adf pair [<host>[:<port>]] [--name <n>] [--yes] | Pair with another computer’s daemon (its 6-digit code); without a host, lists the ADF computers found on this network and your tailnet. Saves the device token |
adf remotes [use <name|local> | forget <name>] | Saved remote daemons; use picks the default |
Daemon
| Command | Description |
|---|
adf daemon [--port <n>] [--host <h>] [--settings <file>] | Run the daemon in the foreground (Ctrl+C stops it; also ADF_DAEMON_PORT, ADF_DAEMON_HOST, ADF_DAEMON_SETTINGS, ADF_USER_DATA_DIR) |
adf daemon start [--port <n>] [--force] | Start it in the background (what adf does on its own); —force skips the ADF Studio check |
adf daemon status | Is it running: pid, uptime, version, log file |
adf daemon stop | Stop it gracefully (agents unloaded, containers stopped) |
adf daemon restart | Stop, then start in the background |
adf daemon logs [-f] [-n <lines>] | Show the background daemon’s log (-f follows) |
adf daemon token | Print this machine’s daemon access token (this machine only; other computers get a device token: daemon devices add / pair) |
adf daemon remote on|off|status | Remote access (live, no restart); also trust|untrust, serve on|off |
adf daemon devices [list|add <label>|revoke <id|label>] | Computers with a device token |
adf daemon pair | Open a pairing window and show its code |
adf daemon --help:
Usage: adf daemon [--port <n>] [--host <h>] [--settings <file>]
adf daemon start [--port <n>] [--force]
adf daemon status | stop | restart [--port <n>]
adf daemon logs [-f] [-n <lines>] [--port <n>]
adf daemon token
adf daemon remote on [--name <name>] | off | status | trust | untrust | serve on|off
adf daemon devices [list | add <label> | revoke <id|label>]
adf daemon pair
Without a subcommand the daemon runs in the foreground (Ctrl+C stops it).
start runs it in the background, like adf does on its own when it needs one;
the log goes to <data dir>/logs/adf-daemon.log. stop is graceful: agents are
unloaded and compute containers stopped.
Every request but GET /health needs the daemon's access token. adf reads it
on this machine by itself (<data dir>/daemon-token, created on first start);
token prints it for a client elsewhere (--token or ADF_DAEMON_TOKEN there).
ADF_DAEMON_TOKEN overrides the file, and is required with a non-loopback
--host (plus ADF_DAEMON_ALLOWED_HOSTS for host names clients use).
Behind a reverse proxy on this host set ADF_DAEMON_BEHIND_PROXY=1: identity
secrets and stop then work only from adf on this host, against its loopback port.
remote on (remote access) does this without any of that: the daemon
listens on this computer's private-network and Tailscale addresses too (no
restart), and each other computer gets its own device token (pair, or
devices add). A device token is as powerful as your owner seed phrase.
See adf daemon remote --help.
Environment: ADF_DAEMON_PORT, ADF_DAEMON_HOST, ADF_DAEMON_SETTINGS,
ADF_USER_DATA_DIR, ADF_DAEMON_PIDFILE, ADF_DAEMON_TOKEN,
ADF_DAEMON_ALLOWED_HOSTS, ADF_DAEMON_BEHIND_PROXY
Options
| Option | Description |
|---|
--url, -u <url> | Daemon URL |
--remote <name> | A daemon saved by adf pair (adf remotes) |
--token <token> | Daemon access token (default: ADF_DAEMON_TOKEN, else the local daemon’s token file) |
--json | JSON output |
--no-daemon | Never start the daemon automatically |
--version, -v | Print the version |
Terminal app options
adf [tui] [options] (adf tui --help). Any of these flags opens the terminal app.
| Option | Description |
|---|
--url, -u <url> | Daemon URL (default ADF_DAEMON_URL, the default remote, or http://127.0.0.1:7385) |
--token <token> | Bearer token (default: ADF_DAEMON_TOKEN, else the local daemon’s token file) |
--view <id> | Start view: chat | files | loops | inspect | fleet | runtime |
--agent <id|handle> | Preselect an agent |
--loop <name> | Preselect one of its loops (default main) |
--theme <name> | adf (dark) | adf-light | adf-contrast | parchment | adf-mono (also ADF_TUI_THEME) |
--mono | No color (also NO_COLOR=1) |
--ascii | ASCII glyphs (also ADF_TUI_ASCII=1) |
--no-alt-screen | Render in the main screen buffer (also turns mouse mode off) |
--mouse | Mouse mode, the default: the wheel scrolls what is under the pointer, a click expands a tool call or thinking, drag selects and copies, right-click pastes (Shift+drag, Option+drag in iTerm2: the terminal’s own selection) |
--no-mouse | The terminal’s own mouse (also ADF_TUI_MOUSE=0, /mouse off; remembered) |
--no-kitty | Never enable the kitty keyboard protocol (Shift+Enter; also ADF_TUI_KITTY=0) |
-h, --help | Show this help |
Environment
| Variable | Used by | Effect |
|---|
ADF_DAEMON_URL | CLI, terminal app | Daemon URL (default http://127.0.0.1:7385); --url wins |
ADF_DAEMON_TOKEN | CLI, terminal app | Daemon access token. Clients: sent as the bearer token (--token wins). Daemon: used instead of the <data dir>/daemon-token file; required with a non-loopback --host |
ADF_NO_AUTOSTART | CLI, terminal app | 1: never start the daemon automatically (same as --no-daemon) |
ADF_REMOTE | CLI, terminal app | A remote saved by adf pair to talk to (same as --remote <name>); local = this computer’s daemon |
ADF_DAEMON_PORT | daemon | Port for adf daemon (default 7385); adf daemon <sub> also uses it when no --url / --port is given |
ADF_DAEMON_HOST | daemon | Bind address for adf daemon (default 127.0.0.1). Non-loopback needs ADF_DAEMON_TOKEN |
ADF_DAEMON_SETTINGS | daemon | Settings file path; its folder is the data dir (token, pid file, logs, identity file). Clients use it to find the local token file |
ADF_USER_DATA_DIR | daemon | Overrides the user data folder that holds the default settings file (shared with ADF Studio) |
ADF_DAEMON_PIDFILE | daemon | Pid file path (default <data dir>/adf-daemon.pid, adf-daemon-<port>.pid for other ports) |
ADF_DAEMON_ALLOWED_HOSTS | daemon | Extra Host names a non-loopback daemon accepts (comma or space separated, name or name:port): the names remote clients or a proxy use |
ADF_DAEMON_BEHIND_PROXY | daemon | 1: a reverse proxy on this host forwards to the daemon. Owner identity secrets and daemon stop then also need <data dir>/daemon-local-proof, which only adf on this host sends: run them on the host against its loopback port |
ADF_OWNER_PASSPHRASE | daemon | Unlocks a passphrase-file owner identity at daemon start (machines without an OS keychain) |
ADF_OWNER_PASSPHRASE_FILE | daemon | Same, read from the first line of this file |
ADF_SECRET_STORE | daemon | file or keychain: where the owner phrase and provider keys are kept (default: the OS keychain when usable, else the passphrase file). Unset or empty: that default. With keychain and no usable keychain, or any other nonempty value, the daemon does not start |
ADF_KEYCHAIN | daemon | 0 or off: never use the OS keychain (conflicts with ADF_SECRET_STORE=keychain) |
ADF_TUI_THEME | terminal app | Same as --theme |
ADF_TUI_ASCII | terminal app | 1: same as --ascii |
ADF_TUI_MOUSE | terminal app | 0: the terminal’s own mouse, 1: mouse mode; wins over the saved /mouse choice |
ADF_TUI_KITTY | terminal app | 0: same as --no-kitty |
ADF_TUI_SHIFT_ENTER | terminal app | 1: treat Shift+Enter as working from the start (your terminal sends it through a keybinding) |
ADF_TUI_FULL_HEIGHT | terminal app | 1: Windows: also use the bottom terminal row |
ADF_TUI_CLIPBOARD | terminal app | osc52: copy only through the terminal (OSC 52), never pbcopy / clip / wl-copy / xclip |
ADF_TUI_PREFS | terminal app | Where sidebar, mouse, stopped-agents filter and welcome choices are kept (default <config dir>/adf-studio/tui-prefs.json); off: session only |
ADF_EDITOR | terminal app | External editor, before VISUAL and EDITOR (default notepad on Windows, else nano, then vi) |
VISUAL | terminal app | External editor when ADF_EDITOR is not set |
EDITOR | terminal app | External editor when ADF_EDITOR and VISUAL are not set |
NO_COLOR | terminal app | Any value: no color (same as --mono); wins over --theme |
TERM | terminal app | dumb: no color |
Terminal app
Views
| Key | View | Id (--view, /view) |
|---|
1 | Chat | chat |
2 | Files | files |
3 | Loops | loops |
4 | Inspect | inspect |
5 | Fleet | fleet |
6 | Runtime | runtime |
Slash commands
Type / in the prompt; Tab completes names and arguments. Commands work from any view (some need a selected agent); the group is the view that owns it, and usually the one it opens. Unknown commands are reported, never sent to an agent.
The selected agent’s registered lambda commands also appear in completion, help and the palette. Built-in names and aliases win; use /command <name> [args] for a collision such as /command status. Add !authorize for one authorized invocation. These commands target the selected loop without a model turn; Enter expands their result in the transcript. See owner slash commands.
Shell (every view)
| Command | Aliases | Description |
|---|
/help | /? | Keys, commands and what loops are |
/quit | /exit /q | Leave the terminal app (agents keep running in the daemon) |
/view <chat|files|loops|inspect|fleet|runtime> | | Switch the main pane |
/agent <handle|id> [loop] | /a | Select an agent, and optionally one of its loops (default main) |
/refresh | /r | Re-read agents, loops and approvals from the daemon |
/theme [adf|adf-light|adf-contrast|parchment|adf-mono|next] | | Pick a color theme (no argument opens the picker) |
/url [daemon-url] [--token <token>] | | Show the daemon URL, or reconnect to another daemon (for a remote daemon pass —token or set ADF_DAEMON_TOKEN) |
/auth | | Provider sign-in: ChatGPT, Grok and Claude status, sign in, sign out |
/login [chatgpt|grok|claude] | | Sign the daemon in to ChatGPT (browser), Grok (device code) or Claude (Claude Code on the daemon host) |
/logout <chatgpt|grok|claude> | | Sign the daemon out of ChatGPT, Grok or Claude (asks) |
/json [on|off] | | Toggle raw JSON in Inspect and Runtime pages |
/sidebar [on|off] | | Show or hide the fleet sidebar (Ctrl+B); remembered across launches |
/mouse [on|off] | | Mouse mode (on, the default: click expands, drag selects + copies, right-click pastes, the wheel scrolls under the pointer; off: the terminal’s own mouse); remembered |
/terminal-setup | /terminal | Shift+Enter for a newline in this terminal: what works now and how to set it up |
/identity [create|restore|unlock|lock] | | Owner identity: status, create, restore from the 12 words, unlock, lock |
/new [name] | | Create a new agent from a template (sealed under your owner identity) |
/web [on|off] | | The daemon’s web server that serves agent websites / APIs: status, start (on), stop (off, asks) |
/open-site [agent] | /site | Open an agent’s website in the browser (starts the web server if it is stopped) |
/copy-site [agent] | | Copy an agent’s website URL |
/welcome | | What ADF is, and the getting-started checklist |
/channels [add [telegram|discord|slack|email|whatsapp] | remove <channel>] | | The selected agent’s channels (Telegram, Discord, Slack, email, WhatsApp): state, add, remove |
/provider [add [preset] | list | remove <id>] | | Model providers: add one with an API key (stored in the daemon’s secret store), list, remove. Subscriptions: /login |
/mcp [add [catalog-name | npm:pkg | python:pkg | https://url] | remove <name> | restart <name> | logs <name>] | | The selected agent’s MCP servers: state, add (catalog, npm, Python, remote URL), restart, credentials, tools, logs, remove |
/templates [template-id] | | Agent templates: list, details, new, duplicate, rename, notes, default, review, reset, delete, edit in $EDITOR |
/skills [add [name | https-url | ./path] | mute|unmute|remove <name> | sources [add|remove <url>]] | | The selected agent’s skills: preview, mute, edit, remove; add from the catalog, a URL or a local folder |
/context [loop] | | Context usage of the selected loop: total vs the auto-compact threshold, what fills it (prompt, files, tools, MCP, conversation), compact now |
Chat view
| Command | Aliases | Description |
|---|
/command <name> [args] | | Run an agent lambda command; add !authorize for one authorized run |
/interrupt | /abort | End the selected loop’s running turn; the loop goes idle and keeps working (Esc) |
/clear | | Clear the selected loop’s history (archived per the agent’s audit settings) |
/compact | | Summarize the selected loop’s history now to free context (not while it is mid-turn) |
/trigger <type> [json-data] | | Fire an ADF event into the selected loop (bypasses trigger config) |
/copy [n] | | Copy the last (or n-th last) reply of the selected loop to the clipboard |
/thinking | | Expand or collapse thinking blocks in chat (t) |
/approve [all|always] | | Approve the selected loop’s pending tool call · all: every gated call (never protection overrides) · always: always approve that tool (asks) |
/reject [feedback] | | Reject the selected loop’s pending tool call; the agent sees the feedback |
Files view
| Command | Aliases | Description |
|---|
/files | | Browse the selected agent’s document, mind and files |
/open <path> | | Open a file of the agent in the viewer (fuzzy path; also document, mind) |
/edit <path> | | Edit a file of the agent in $VISUAL/$EDITOR, then confirm the write |
/doc | /document | Show the agent document |
/mind | | Show the agent’s mind (its memory; inner loops such as a consolidator tend it) |
/new-file [path] | | Create a file in the agent (opens $EDITOR) |
/rm <path> | | Delete a file of the agent (asks first) |
/mv <from> <to> | | Rename or move a file of the agent |
Loops view
| Command | Aliases | Description |
|---|
/loop [<name> | new [template | <name> <goal…>] | edit|rm|on|off <name> | send <name> <msg…> | chat <name> | schedule <name>] | | Loops: switch, create (templates), edit, enable/disable, delete, message or schedule an inner loop |
/main | | Back to the main loop |
/loops | | Loops manager: this agent’s loops, timers, triggers and history |
/timers [all] | | Timers of this agent, or all agents’ upcoming timers |
/timer add [--loop <name>] | edit <id> | rm <id> | | Create, edit or delete a timer (a timer with a loop runs that loop on a schedule) |
/triggers | | Triggers of this agent: enable/disable, edit targets and the loop each target wakes |
/history [loop] | | Browse a loop’s persisted history (tokens per entry, filters) |
Inspect view
| Command | Aliases | Description |
|---|
/inspect [diag|settings|config|usage|mcp|channels|identity|logs|tables|tasks|events] | | Inspect the selected agent: status, settings, config, usage, MCP, channels, identities, logs, tables, its events |
/model [[provider/]model | inherit] | | Change the selected agent › loop’s model (main: the agent config; an inner loop: its override, inherit clears it). No argument: picker |
/config [edit] | | The selected agent’s config (Inspect › Config); edit opens it in $EDITOR |
/tasks [pending|active|all] | | The selected agent’s tasks (Inspect › Tasks): approvals and async tool calls; y/n/a act on a call waiting for approval |
/instructions [edit] | | Edit the selected agent’s instructions (a dialog; edit opens $EDITOR) |
/tools [filter] | | The selected agent’s tools: enable, show, require approval, lock (built-in and MCP) |
/compaction [tokens|default] | | Compaction threshold of the selected agent › loop (80000, 80k; default = main’s default / an inner loop inherits). No argument: every loop |
/media [<files> <MB>|default] | | Request media limit of the selected agent: most files and MB one request carries (default 20, 20). Past either, the oldest are left out until half remains |
Fleet view
| Command | Aliases | Description |
|---|
/agents [running|all | interrupt|abort <agent> [loop] | refresh] | | Fleet overview; running/all: hide or show stopped tracked agents (remembered); interrupt an agent’s running turns (abort = hard stop until reload); refresh |
/start [agent] | | Start an agent (runs its startup turn when it starts active); a stopped tracked agent is loaded first |
/stop [agent] | /unload | Stop and unload an agent and all its loops (asks first) |
/load [path.adf] [--review] [--start] | | Load an .adf into the daemon (no path: file picker with Tab completion) |
/switch <agent>[/loop] | /sw | Jump to an agent (fuzzy) or one of its loops and open the chat |
/autostart [dir…] | | Scan tracked directories and start every reviewed autostart agent (asks first) |
/track [dir] | | Track a folder of agents (its reviewed autostart agents load now and at every daemon start); no dir: folder picker |
/untrack <dir> | | Stop tracking a folder (asks; files are not touched; optionally unload its agents) |
Runtime view
| Command | Aliases | Description |
|---|
/runtime [status|folders|identity|auth|providers|usage|network|compute|mcp|channels|settings|events] | | The daemon: status, owner identity, sign-in, providers, usage, network, compute, MCP, channels, settings, every agent’s events |
/status | | Daemon health, version, uptime and every agent’s loops (Runtime › Status) |
/usage | | Token usage across agents, by model (Runtime › Usage) |
/providers | | LLM providers and which agents use them (Runtime › Providers) |
/network | | Mesh, LAN and WebSocket status; mesh and server on/off (Runtime › Network) |
/compute | | Container runtime and containers (Runtime › Compute) |
/settings | | Daemon settings (secrets redacted) (Runtime › Settings) |
/events [type-filter…] [--agent] [--loop] [--all] | | Every agent’s live umbilical events (tail -f), e.g. /events tool. -turn.delta —agent (Runtime › Events) |
Palette actions
Ctrl+K (or Ctrl+P, or : outside the prompt) searches these, plus views, agents, loops, slash commands and recent files.
Shell (every view)
| Action | Group | Shortcut |
|---|
| Refresh from daemon | Actions | |
| Keys & commands | Actions | ? |
| Next color theme | Actions | |
| Toggle raw JSON in Inspect and Runtime | Actions | |
| Show / hide the sidebar | Actions | Ctrl+B |
| Mouse mode on / off | Actions | |
| Terminal setup: Shift+Enter newline | Actions | |
| Quit TUI (agents keep running) | Actions | Ctrl+C |
| Owner identity | Actions | |
| Create owner identity | Actions | |
| Restore owner identity from seed phrase | Actions | |
| Unlock owner identity | Actions | |
| New agent | Actions | n |
| Provider sign-in | Actions | |
| Sign in to ChatGPT | Actions | |
| Sign in to Grok | Actions | |
| Sign in to Claude | Actions | |
| Open agent website | Web | w |
| Copy agent website URL | Web | Shift+w |
| Start web server | Web | |
| Stop web server… | Web | |
| MCP servers of this agent | Actions | |
| Add an MCP server | Actions | |
| Welcome & getting started | Actions | |
| Channels of this agent | Actions | |
| Add Telegram channel | Actions | |
| Add Discord channel | Actions | |
| Add Slack channel | Actions | |
| Add Email channel | Actions | |
| Add WhatsApp channel | Actions | |
| Connect a model provider | Actions | |
| Agent templates | Actions | |
| Skills of this agent | Actions | |
| Add a skill | Actions | |
| Context usage | Actions | |
Chat view
| Action | Group | Shortcut |
|---|
| Interrupt running turn | Chat | Esc |
| Next loop | Chat | Shift+→ |
| Previous loop | Chat | Shift+← |
| Toggle thinking blocks | Chat | t |
| Copy last reply | Chat | c |
| Fire a trigger into this loop… | Chat | |
| Compact this loop’s history now | Chat | |
| Clear this loop’s history… | Chat | |
| Approve the pending tool call | Chat | |
| Approve all pending tool calls | Chat | |
| Always approve this tool… | Chat | |
| Reject the pending tool call with feedback… | Chat | |
Files view
| Action | Group | Shortcut |
|---|
| Show agent document | Files | |
| Show agent mind | Files | |
| Edit agent mind in $EDITOR | Files | |
| New file… | Files | |
| Browse files | Files | |
| Show inbox | Files | |
| Show outbox | Files | |
| Show meta | Files | |
Loops view
| Action | Group | Shortcut |
|---|
| New inner loop… | Loops | |
| New loop from template: Memory consolidator | Loops | |
| New loop from template: Researcher | Loops | |
| New loop from template: Critic / reviewer | Loops | |
| New loop from template: Reflector | Loops | |
| Manage loops | Loops | |
| Schedule the selected loop… | Loops | |
| Timers | Loops | |
| Upcoming timers (all agents) | Loops | |
| Triggers | Loops | |
| Loop history | Loops | |
| Edit the selected loop… | Loops | |
Inspect view
| Action | Group | Shortcut |
|---|
| Agent settings | Inspect | |
| Edit agent instructions | Inspect | |
| Agent tools | Inspect | |
| Compaction threshold | Inspect | |
| Show tasks | Inspect | |
| Change model | Inspect | |
| Edit agent config in $EDITOR | Inspect | |
| Events of the selected loop only | Inspect | |
| Inspect: Status | Inspect | |
| Inspect: Settings | Inspect | |
| Inspect: Config | Inspect | |
| Inspect: Usage | Inspect | |
| Inspect: MCP | Inspect | |
| Inspect: Channels | Inspect | |
| Inspect: Identities | Inspect | |
| Inspect: Logs | Inspect | |
| Inspect: Tables | Inspect | |
| Inspect: Tasks | Inspect | |
| Inspect: Events | Inspect | |
Fleet view
| Action | Group | Shortcut |
|---|
| Track a folder… | Fleet | f |
| Untrack folder… | Fleet | |
| Load agent file (.adf)… | Fleet | o |
| Start selected agent | Fleet | s |
| Stop + unload selected agent… | Fleet | x |
| Interrupt running turns of selected agent… | Fleet | a |
| Hard abort selected agent’s running loops (stopped until reload)… | Fleet | |
| Autostart agents from tracked directories… | Fleet | Shift+a |
| Hide / show stopped agents (tracked, not loaded) | Fleet | Shift+h |
| Refresh fleet (agents, timers, inbox, runtime) | Fleet | r |
| Open selected agent + loop in chat | Fleet | Enter |
Runtime view
| Action | Group | Shortcut |
|---|
| Runtime: Status | Runtime | |
| Runtime: Folders | Runtime | |
| Runtime: Identity | Runtime | |
| Runtime: Sign-in | Runtime | |
| Runtime: Providers | Runtime | |
| Runtime: Usage | Runtime | |
| Runtime: Network | Runtime | |
| Runtime: Compute | Runtime | |
| Runtime: MCP | Runtime | |
| Runtime: Channels | Runtime | |
| Runtime: Settings | Runtime | |
| Runtime: Events | Runtime | |
Keys
The same lists as ? / /help in the app.
Global
| Keys | Action |
|---|
Ctrl+K Ctrl+P | Command palette (also : when the prompt is not focused) |
Tab Shift+Tab | Cycle focus: sidebar → main → prompt (a hidden sidebar is skipped); from the tab bar Tab goes to the first pane |
Ctrl+B | Hide / show the fleet sidebar: full-width chat (also /sidebar; remembered) |
1-6 Alt+1-6 | Switch view. The selected agent: 1 Chat 2 Files 3 Loops 4 Inspect · the app: 5 Fleet (every agent) 6 Runtime (the daemon) (bare digits outside the prompt) |
Shift+← Shift+→ | Previous / next loop of the selected agent, from anywhere (also Ctrl+←/→, except in a prompt with text) |
Alt+← Alt+→ | In the prompt: word left / right (also Alt+B / Alt+F, and Ctrl+←/→ while there is text) |
Home End | In the prompt: line start / end (also Ctrl+A / Ctrl+E, Cmd+← / Cmd+→) |
Ctrl+Backspace Alt+Backspace Ctrl+W | In the prompt: delete the word before the caret (Option+Backspace on a Mac) |
Ctrl+Delete Alt+d | In the prompt: delete the word after the caret |
Cmd+Backspace Ctrl+U | In the prompt: delete to the line start (Terminal.app sends nothing for Cmd+Backspace: use Ctrl+U) |
/ | Slash command (focuses the prompt) |
? | This help (when the prompt is not focused) |
Esc | In order: close the dialog or completion menu · end a view’s own mode (filter, search, viewer, a selected chat item) · (chat) interrupt the running turn · then focus the tab bar (from the sidebar, the main pane, or an empty prompt) |
← → | Files, Loops, Inspect, Runtime: the view’s tabs (from its lists; not while typing a filter) |
Enter Backspace | In a list: open the item (file, message, entry, task, row, event) · back one level (Esc too). Backspace never deletes |
Esc Esc | Clear a prompt that has text (the next Esc goes to the tab bar) |
Ctrl+C | Cancel dialog · clear the prompt · press twice to quit (agents keep running) |
Enter | Send the prompt (to the selected agent › loop unless it starts with /). With a completion menu open: take the highlighted item; a /command with all its required arguments runs, else it is inserted to type them |
Shift+Enter Alt+Enter Ctrl+J | Newline in the prompt (or end a line with ). Shift+Enter needs a terminal that tells it apart: /terminal-setup |
Ctrl+↑ Ctrl+↓ | Prompt history (per agent › loop in Chat); ↑ ↓ too on the first / last line of text |
↑ ↓ | Chat, empty prompt: select the previous / next transcript item (Enter expands, Esc lets go; PgUp PgDn Home End scroll) · otherwise move in the text / completion menu |
Mouse | Mouse mode (default): the wheel scrolls what is under the pointer; click selects / expands a tool call; drag selects + copies (double-click word, triple-click line); right-click pastes. Shift+drag (Option+drag iTerm2, Fn+drag Terminal.app): the terminal’s own selection. /mouse off: the terminal’s mouse |
w | Fleet, Inspect › Status, Chat transcript: open the agent’s website (W copies the URL) · on the tab bar: web server on / off |
Tab | In the prompt with a menu open: complete /commands, their arguments, @files (Chat) without running |
Tab bar
Esc focuses the view tabs in the header once nothing else takes it.
| Keys | Action |
|---|
← → | Previous / next view: switches as you move (h l too; Home / End first / last). Left of 1 Chat is the agent label |
Enter | On the agent label: switch agent / loop (also a click on it in mouse mode) |
Enter ↓ | Into the view (Chat: its prompt) |
1-6 | Jump straight into a view |
Tab Shift+Tab | Back into the panes (first / last) |
w | Web server on / off (stopping asks); also a click on the header’s web badge in mouse mode, /web on|off |
| Keys | Action |
|---|
↑ ↓ | Move; selects that agent › loop everywhere (PgUp PgDn Home End too) |
→ ← | Expand an agent into its loops · collapse / back to the agent |
Space | Expand or collapse the agent |
Enter | Open the agent › loop in Chat |
a-z | Type to filter agents and loops (Backspace edits, Esc clears) |
Dialogs
| Keys | Action |
|---|
y Enter | Confirm |
n Esc | Cancel |
1 Chat
| Keys | Action |
|---|
Enter | Send to the selected loop (queued while it runs) |
Shift+← Shift+→ | Previous / next loop tab (also Ctrl+←/→ on an empty prompt; ← → in the transcript) |
↑ ↓ | Select the previous / next transcript item (message, reply, thinking, tool call, notice, approval, marker): from an empty prompt or the transcript (j k). An item taller than the view scrolls inside first |
Enter Space | Expand / collapse the selected item (tool call input + result, thinking, context, approval, wake). Mouse mode: click the item |
Esc | Let go of the selected item (before interrupting) |
PgUp PgDn | Scroll by pages (also while typing); the mouse wheel scrolls by lines |
Home End | With an empty prompt: top, loading older history · follow live (g G in the transcript) |
Ctrl+↑ Ctrl+↓ | Prompt history of this agent › loop (↑ ↓ too once the prompt has text, on its first / last line) |
Shift+Enter Alt+Enter | Newline in the prompt (Ctrl+J, or \ then Enter; /terminal-setup for Shift+Enter) |
t | Expand / collapse all thinking |
c | Copy the selected item, else the last reply |
w | Open the agent’s website (when it serves one; starts the web server if stopped) |
Esc | Interrupt this loop’s running turn: it goes idle and keeps working (not a stop) |
Esc Esc | Clear the prompt |
Tab Shift+Tab | Leave the prompt for the transcript / sidebar (Esc never does) |
@ | In the prompt: complete an agent file path (Tab) |
Approvals and questions (the card under the transcript):
| Keys | Action |
|---|
y | Approve: press twice, or once on a selected approval · resume a suspended agent. Mouse mode: click the card’s buttons |
a | Always approve this tool for the agent (twice, then confirm; it won’t ask again). Not offered for protection overrides: those are one-time only · on a question card: answer it in the prompt |
x | On a question card: skip the questions not answered yet and send the answers |
n | Reject · shut a suspended agent down (asks) |
f | Reject with feedback: the agent sees your text |
v | Full details of the tool call |
/approve | /approve [all|always] · /reject [feedback]: the same from the prompt |
2 Files
Files tab (tree):
| Keys | Action |
|---|
↑ ↓ | Move; the viewer previews the highlighted entry |
Enter | Open the file in the viewer · open / close a folder |
Backspace | Close the folder · go to the parent folder (never deletes) |
← → | Tabs: Files · Inbox · Outbox · Meta |
Space | Toggle a folder |
/ | Filter by path (fuzzy) |
e | Edit in $EDITOR, then confirm the write (diff shown) |
o | Open in the default app (xlsx, images, pdf, …) |
s | Save the default-app copy back (asks) |
n | New file |
m | Rename / move (a folder moves with its files) |
d Delete | Delete (asks) |
p | Protection: none → read_only → no_delete |
a | Toggle authorized |
r | Reload |
Viewer:
| Keys | Action |
|---|
Backspace Esc | Back to the list (the entry stays selected) |
↑ ↓ | Scroll (j k; PgUp PgDn Space page; g G top/bottom) |
/ | Search, then n N next / previous hit |
e | Edit this file |
o s | Open in the default app · save that copy back |
Inbox · Outbox · Meta (read-only):
| Keys | Action |
|---|
Enter | Read the message / entry (Backspace or Esc: back) |
/ | Filter |
f | Status filter (inbox, outbox) |
r | Reload |
3 Loops
| Keys | Action |
|---|
← → | Tabs: Loops · Timers · Triggers · History |
Loops tab:
| Keys | Action |
|---|
↑ ↓ | Select a loop (also selects it for Chat) |
Enter | Chat with the loop |
n | New inner loop: memory consolidator, researcher, critic, reflector or blank |
e | Edit goal, tools, model, flags (diff + confirm) |
x | Enable / disable |
s | Send a one-off message |
t | Schedule: a timer that wakes this loop |
c | Clear its history (asks) |
d Delete | Delete (asks; history archived to the audit log) |
h | Its history |
r | Refresh |
Timers tab:
| Keys | Action |
|---|
n | new |
Enter e | edit |
d | delete |
f | all agents |
r | refresh |
o | select that agent (all-agents list) |
Triggers tab:
| Keys | Action |
|---|
x | enable/disable |
Enter e | edit targets |
r | refresh |
History tab:
| Keys | Action |
|---|
Enter | open entry |
/ | filter text/tool |
f | role |
< > | older/newer |
l | next loop |
c | clear |
L | previous loop |
i | entry details |
r | reload |
Loop and timer forms:
| Keys | Action |
|---|
↑ ↓ | Field |
← → | Change a choice / toggle |
Ctrl+O | Open the goal in $EDITOR |
Ctrl+S | Review, then Enter / y applies |
Esc | Cancel (from review: back to the form) |
4 Inspect
| Keys | Action |
|---|
← → | The selected agent’s tabs: Status Settings Config Usage MCP Channels Identities Logs Tables Tasks Events |
Other tabs:
| Keys | Action |
|---|
e | Config: edit in $EDITOR (validated, changed keys confirmed; also /config edit) |
w | Status: open the agent’s website (W copies its URL) |
Enter | Logs / Tables: details · row |
f | Logs: follow |
n p | Tables: next / previous page |
r | Reload |
Settings (instructions, tools, compaction, autonomy, messaging, host access):
| Keys | Action |
|---|
Enter Space | Open the dialog (Instructions, Tools, Compaction, Visibility, Send mode) or flip the switch; Autonomous and Host access ask first |
e | Edit the instructions in $EDITOR (also /instructions edit) |
l | Lock / unlock that section for the agent (its sys_update_config; you can always change it) |
r | Reload config and context use |
Tools dialog (/tools [filter]):
| Keys | Action |
|---|
Space Enter | Enable / disable (on a group header: the whole group, locked tools skipped) |
v | Show / hide in the agent’s tool list (hidden tools stay callable from code) |
r | Require approval (restricted): an LLM call waits for you |
l | Lock / unlock a built-in tool for the agent (locked: enable and show are fixed until unlocked) |
/ | Filter by name or description (Enter keeps it, Esc clears) |
Instructions dialog (/instructions):
| Keys | Action |
|---|
Ctrl+S | Save (re-reads first; asks if they changed on the daemon meanwhile) |
Enter | New line |
Ctrl+O | Continue in $EDITOR |
Esc | Cancel (asks when there are unsaved changes) |
Compaction dialog (/compaction [tokens|default]):
| Keys | Action |
|---|
Enter e | Set the threshold of main or an inner loop (80000 or 80k; default / inherit resets; empty keeps it) |
d | Back to the default (main) or main’s threshold (an inner loop) |
c | Compact that loop now (asks) |
Tasks (approvals and async tool calls; also /tasks):
| Keys | Action |
|---|
f | Filter: pending · active (+ running) · all |
/ | Search tool, status, reason, loop, origin, args |
Enter Esc | Detail (args, result, error, timestamps) · back |
y n | Approve · reject the highlighted call waiting for approval |
a | Always approve its tool (confirmed; not offered for one-time approvals) |
N | Reject with feedback (the agent sees the text) |
A | Approve all waiting calls (confirmed; protection overrides skipped) |
Events (this agent’s live umbilical events; every agent: Runtime › Events):
| Keys | Action |
|---|
↑ ↓ | Move; ↑ stops following (the wheel scrolls too) |
End | Follow the newest again (also G) |
t | Type filter, e.g. tool. -turn.delta |
l | All loops / the selected loop |
/ | Search text (Esc clears) |
Space | Pause / resume |
f | Follow on / off |
Enter | Full event |
c | Clear |
x | Reset filters |
5 Fleet
| Keys | Action |
|---|
n | New agent from a template (sets up your owner identity first if needed) |
i | Owner identity: status, create, restore, unlock (I hides the banner) |
c r u | With no agents and no ready identity: create, restore, unlock the owner identity |
↑ ↓ | Select an agent (j k) |
← → | Select one of its loops (h l) |
Enter | Open the agent › loop in Chat |
s | Start the agent (a stopped one: load it into the daemon, then start it; one that needs review: review it first) |
H | Hide / show stopped agents: tracked agents that are not loaded (also /agents running|all; remembered) |
x | Stop and unload the agent (asks; the .adf is kept) |
a | Interrupt the turns running now (asks) |
w | Open the agent’s website (starts the web server if it is stopped) |
W | Copy the agent’s website URL |
o | Load an .adf (Tab completes, ^R require review, ^S start after load) |
f | Track a folder of agents (Tab completes folders): its reviewed autostart agents load now and at every daemon start · /untrack stops |
A | Autostart: scan tracked directories (asks) |
r | Refresh agents, timers, inbox and the daemon line |
6 Runtime
| Keys | Action |
|---|
← → | The daemon’s tabs: Status Folders Identity Sign-in Providers Usage Network Compute MCP Channels Settings Events |
Pages:
| Keys | Action |
|---|
r | Reload the page |
Enter | Identity: the identity dialog · Sign-in: the sign-in dialog |
c r u | Identity: create / restore / unlock (when offered) |
Enter | Folders: the selected folder’s agents (loaded, needs review, errors): review + accept + load, or load |
a d | Folders: track a folder (its reviewed autostart agents load now) / stop tracking the selected one (asks; files untouched, optionally unload its agents) |
m | Network: mesh on / off (asks) |
s | Network: web server (serves agent sites, APIs, mesh delivery) start / stop (stopping asks) |
R | Network: restart the web server (asks) |
Events (every agent’s live umbilical events):
| Keys | Action |
|---|
↑ ↓ | Move; ↑ stops following (the wheel scrolls too) |
End | Follow the newest again (also G) |
t | Type filter, e.g. tool. -turn.delta |
a | All agents / the selected agent (Runtime only) |
l | All loops / the selected loop |
/ | Search text (Esc clears) |
Space | Pause / resume |
f | Follow on / off |
Enter | Full event |
c | Clear |
x | Reset filters |
Legend
| Glyph | Meaning |
|---|
● (*) | Running: a turn is in progress (agent: in any of its loops; loop: in that loop) |
○ (o) | Idle: loaded and waiting for work |
✗ (x) | Disabled loop, or an agent in error / stopped |
↻ (@) | A loop: ↻main is the agent itself, the rest are inner (side) loops |
!N | N approvals or questions waiting for you |
«N | N unread inbox messages |
in 1h | Next timer wake of that loop |
● live / offline | Header: live event stream to the daemon, or the daemon is not answering |
● web :7295 / ○ web off | Header: the web server that serves agent websites / APIs (w on the tab bar, /web on|off) |