Docs / Autopilot / Commands and shortcuts

Commands and keyboard shortcuts

Reference for Autopilot slash commands, keyboard shortcuts, CLI subcommands, command-line flags and exit codes.

Slash commands

Type / at the prompt to open the command list. /help shows the same list grouped by category.

Session

Command Effect
/resume Pick a past conversation to restore
/compact Summarize older turns to free context
/clear Clear the current conversation
/fresh Start a fresh session with a summarized plan or continuation context
/checkpoint [label] Save the current point in the session
/checkpoint list List checkpoints in the current session
/now Run your latest queued message now instead of after the current turn
/good, /bad [reason] Rate the last turn. Recorded in the session trace when tracing is on.
/exit Exit Autopilot

Model and provider

Command Effect
/model Open the model picker for the active provider
/model <id>, /model list [filter] Set a model, or list models
/provider List providers and where each key comes from
/provider <id> [key] Switch provider
/key Show the active provider's key status (and spend and credit on OpenRouter)
/key set <key>, /key clear Replace or remove the key
/logout Clear stored credentials
/reasoning Show or hide model reasoning

Modes and subagents

Command Effect
/settings Show feature settings
/settings modes on|off Turn plan, edit and auto modes on or off (off by default)
/mode edit|plan|auto Switch mode (when modes are on). Also /plan, /edit, /auto.
/subagents List running subagents
/subagents cancel <n>|all Stop one subagent, or all, without interrupting the turn
/subagents off|suggest|auto Set the delegation policy (default auto)
/subagents stats Subagent usage over the last 7 days
/agents, /jobs Open the activity panel for subagents and background jobs
/jev [yes|choose|score] Ask Jev for a typed, scoped one-shot decision

Project and tools

Command Effect
/init Scan the repository and write an AGENTS.md project context file
/memory Show memory stats
/memory on|off Enable or disable memory
/memory clear Wipe memories for this repository
/memory search <query> Search stored memories
/skills list|add|edit|delete|enable|disable Manage skills
/command create|edit|delete|list Manage custom slash commands
/mcp list, /mcp reload List or reconnect MCP servers
/lsp config|list|reload|scope Manage language servers
/hooks [list|recommended|enable <id>|disable <id>|path|reload] Manage hooks
/trust [yes] Review and trust this project's hooks and language servers
/shell [auto|bash|cmd|powershell|<path>] Show or set the shell for the bash tool

Cost and info

Command Effect
/cost Cost by session, day, month and all time
/cost on|off Turn cost attribution by task type on or off
/cost verify Check recorded tokens and cost against OpenRouter's generation records
/update Check for updates
/changelog-image [owner/repo] [days] Generate a changelog image from merged pull requests
/hello Send a voice note to the creator
/help List all commands

Run a shell command yourself

Type ! followed by a command to run it in your own terminal inside the session:

! gh auth login

The command gets the real TTY, so logins, password prompts and interactive programs work. The output is added to the conversation. The agent's own bash tool has no terminal, so Autopilot suggests ! <command> when it needs you for a login, a sudo password, an SSH passphrase or a secret. On macOS and Linux the output is captured with script(1); on other platforms the command runs but its output isn't captured.

Keyboard shortcuts

Shortcut Action
Ctrl+C / Esc Interrupt the current turn when busy; exit when idle
Ctrl+R Toggle reasoning display
Ctrl+O Toggle verbose tool output
Shift+Tab Cycle mode (edit, plan, auto). Only when modes are on.
↑ / ↓ Walk prompt history
@ Mention a file
/ Open the command list

In a permission prompt, use ↑/↓ or j/k to move, and Alt+1, Alt+2, Alt+3 to choose.

Interrupting a turn stops it and any subagents it started in the foreground. Background subagents keep running; stop them with /subagents cancel <n>.

CLI subcommands

Command Effect
autopilot Start the interactive UI
autopilot resume [id] Reopen a saved conversation, or pick one from this directory
autopilot auth <provider> [key] Connect a provider and make it active
autopilot auth use <provider> Make a provider active
autopilot auth status List providers, the active one, and where each key comes from
autopilot trust [dir] [-y] Review and trust a project's hooks and language servers
autopilot logs path|dir|prune Inspect structured logs
autopilot history Show where sessions and traces live and their size
autopilot history prune --older-than <days> Delete sessions and traces not touched in that many days
autopilot cost Cost attribution by task type (-d, -w, -m, -s <id>, -c <category>, --json, --verify)
autopilot eval run, autopilot eval compare Run and compare evals. See Evals.
autopilot serve [--port <n>] [--hostname <host>] Start the headless HTTP server (default 127.0.0.1:4096). See Running unattended.
autopilot remote add <url>, remote list, remote remove <name> Pair with Autopilot servers
autopilot runs [--on <server>] List tasks on your servers
autopilot runs watch <run>, runs cancel <run> Follow or stop a run
autopilot pull <run> Check out a run's branch in this repository

Flags

Flag Effect
-p, --print <prompt> One-shot mode: send the prompt, stream the reply to stdout, exit
-m, --model <id> Model id for the active provider
-c, --continue Continue the most recent session in this directory
-S, --session <id> Resume a specific session
--mode <mode> interactive (default), print or rpc
--format <mode> Print mode output: text (default), json or stream-json
-f, --file <path> Attach files to the prompt. Repeatable, supports globs. Print mode only.
--dir <path> Run in another directory. Print mode only.
--title <title> Override the session title. Print mode only.
--dangerously-allow-all Approve every permission prompt. Print mode only.
--reasoning, --thinking Include reasoning in the output. Print mode only.
--continue-on-limit Reset the tool-call counter and continue when the 200-call limit is hit. Print mode only.
--max-total-tool-iterations <n> Hard ceiling on tool iterations per turn across resets (default 1000)
--max-input-tokens <n> Cumulative prompt token budget; exits 42 when exhausted. Print mode only.
--attach <url> Attach to a running autopilot serve instance. Print mode only.
--emit-events Emit Camouflage NDJSON events to stdout. Requires -p.
--multi-turn With --emit-events: keep reading stdin for follow-ups
--on <server> Run the task on a paired server instead of this machine
--repo <owner/name> With --on: the GitHub repository (default: this checkout's origin)
--detach With --on: start the task and return without streaming
--local Run on this machine (the default)
--version Print the version 
autopilot -p "summarize PLAN.md"
autopilot -p "fix the lint errors" --dangerously-allow-all
autopilot -p "review this diagram" -f docs/arch.png --format json

Exit codes

Code Meaning
0 Success
1 Error
2 The terminal UI could not start, or autopilot serve found no provider credentials
42 --max-input-tokens budget exhausted
43 The agent loop stopped because every tool call was blocked twice
44 A UserPromptSubmit hook blocked the prompt