Commands Reference

All commands start with / and can be tab-completed.

Session commands

/help

Show the list of available commands.

/help

/new

Start a new session. Resets all messages, token counts, input history, checkpoints, and the redo stack. If you were in a git repo, old checkpoint refs are cleaned up.

/new

/sessions

List up to 10 recent sessions for the current working directory. Each entry shows the session ID prefix (8 characters), title, time since last update, and message count.

/sessions

Example output:

  a1b2c3d4  Fix login bug                2h ago  12 messages
  e5f6g7h8  Refactor auth module         1d ago  34 messages

/continue

Resume a previous session by its ID prefix. Loads all messages back into the conversation and restores token counts.

/continue <id-prefix>

No argument: behaves the same as /sessions (lists recent sessions).
With argument: finds a session whose ID starts with the given prefix and resumes it. If no match is found, prints an error.

/branch

Fork the conversation at the current point. Creates a new session that branches from the current message, letting you explore an alternative direction without losing the original conversation.

/branch

TUI only.
The new branch becomes the active session.

/rename

Rename the current session.

/rename <new title>

TUI only.
Updates the session title shown in /sessions output.

/resume

Alias for /continue. Identical behavior.

/resume <id-prefix>

/config

Manage preferences and configuration. See also Configuration for full details.

/config show

Display all preference keys, their current values, and defaults.

/config show

/config models

Show available models and pricing.

/config models

/config tools

Show available agent tools.

/config tools

/config daemon

Show daemon/network settings (bind address).

/config daemon

/config hub

Show hub coordinator settings (bind address, auth token).

/config hub

/config node

Show node registration settings (hub URL, node token, node name).

/config node

/config theme

Show theme settings.

/config theme

/config mobile

Show mobile model picker settings. The mobile_models key must be set by editing config.json directly.

/config mobile

/config set

Set a preference value.

/config set <key> <value>

Provider keys

Key	Description
`anthropic.api_key`	Anthropic API key
`openai.api_key`	OpenAI API key
`mistral.api_key`	Mistral API key
`grok.api_key`	Grok / xAI API key
`zai.api_key`	Z.AI API key
`google.api_key`	Google AI API key
`fireworks.api_key`	Fireworks AI API key
`deepinfra.api_key`	DeepInfra API key
`ollama.url`	Ollama server URL (default: `http://localhost:11434`)
`brave.api_key`	Brave Search API key (for web search tool)
`textbelt.api_key`	Textbelt API key (for SMS tools)

Model

Key	Description
`model`	Default model (e.g. `claude-sonnet`, `gpt-4o`, `claude-opus`)
`model.compact`	Model used for smart compaction
`model.title`	Model used for auto-titling sessions
`model.tags`	Model used for tagging
`model.consult`	Model used for second opinion consults (e.g. `openai/gpt-4o`)

Footer display

Key	Type	Default	Description
`footer.tokens`	bool	`true`	Show token count in footer
`footer.cost`	bool	`true`	Show estimated cost in footer
`footer.cwd`	bool	`true`	Show working directory in footer
`footer.session`	bool	`true`	Show session info in footer
`footer.keybindings`	bool	`true`	Show keybinding hints in footer
`footer.emoji`	string	(none)	Footer emoji (use `/emoji` or set by name e.g. `devil`)
`show_diffs`	bool	`true`	Show inline file diffs

Boolean values accept: true/false, on/off, yes/no, 1/0 (case-insensitive).

Theme

Key	Description
`theme.spinner`	Activity spinner style in the TUI status line and background-job indicator

Daemon

Key	Description
`daemon.bind_address`	Network interface to bind (`localhost`, `0.0.0.0`, or specific IP)
`daemon.auth_token`	Auth token for daemon connections

Hub

Key	Description
`hub.bind_address`	Hub bind address (when running `muxd-hub`)
`hub.auth_token`	Bearer token for hub authentication

Node

Key	Description
`hub.url`	Hub URL this daemon registers with
`hub.node_token`	Token used to authenticate with the hub
`hub.node_name`	Display name for this node in the hub

Tools

Key	Description
`tools.disabled`	Comma-separated list of tools to disable (e.g. `bash,file_write`)
`tools.ask_user`	Enable/disable the ask_user tool (default: `true`)
`permissions.mode`	Tool approval mode (`ask`, `accept-edits`, `full-auto`)
`scheduler.allowed_tools`	Comma-separated list of tools scheduled jobs may run

Examples:

/config set anthropic.api_key sk-ant-...
/config set model claude-opus
/config set model.consult openai/gpt-4o
/config set footer.cost false
/config set theme.spinner mini-dot
/config set permissions.mode accept-edits
/config set daemon.bind_address 0.0.0.0
/config set tools.disabled bash,file_write

/config reset

Reset all preferences to their default values.

/config reset

Config and mode commands

/emoji

Open the interactive emoji picker to choose a footer emoji. The selected emoji appears before "muxd" in the footer and persists across sessions.

/emoji

TUI only.
Also accessible via /config → theme → footer.emoji → Enter.
Alternatively, set by name: /config set footer.emoji devil

/spinner

Open the spinner style picker and preview available activity spinner styles.

/spinner

TUI only.
Persists to theme.spinner (same as /config set theme.spinner <style>).

/permissions

Set tool approval behavior for shell and file-edit operations.

/permissions
/permissions <ask|accept-edits|full-auto>

Modes:

Mode	Behavior
`ask`	Prompt before bash and file edits (default)
`accept-edits`	Auto-approve file edits in the project, still ask for bash
`full-auto`	Auto-approve most actions, but still blocks sensitive paths (for example `.env`, credentials, and SSH keys)

You can also configure this with /config set permissions.mode <mode>.

Editing and transfer commands

/undo

Undo the last agent turn by restoring the working tree to the checkpoint taken before that turn.

/undo

Requires git (must be inside a git repository).
Cannot undo while the agent is running.
Prints an error if there are no checkpoints to undo.
Multiple /undo commands walk back through the checkpoint stack.

See Undo / Redo for a detailed explanation.

/redo

Re-apply the last undone agent turn.

/redo

Only available after an /undo.
The redo stack is cleared whenever the agent runs a new turn.
Same requirements as /undo (git, not during agent loop).

/clear

Clear the chat display. Resets displayed messages and token counters, but does not end the session. Messages remain persisted in the database.

/clear

Network and automation commands

/schedule

Manage scheduled tool jobs and agent tasks.

/schedule list                                           # list all jobs
/schedule cancel <id>                                    # cancel a job
/schedule add <tool> <HH:MM|RFC3339> <json>              # schedule a single tool call
/schedule add-task <HH:MM|RFC3339> <prompt> [--daily|--hourly]  # schedule a full agent task

/schedule add-task

Schedule a multi-step agent task. Unlike /schedule add (which fires a single tool call), add-task spawns a full agent loop at the scheduled time with all tools available.

/schedule add-task 16:00 Search for Go articles and summarize the top 5
/schedule add-task 08:00 Summarize my recent git commits --daily
/schedule add-task 2026-03-01T10:00:00Z Generate a weekly project report --hourly

Time accepts HH:MM (local, auto-advances to next day if past) or RFC3339.
Append --daily or --hourly for recurring tasks (default: once).
The scheduled agent runs headless — ask_user is disabled.
In /schedule list, agent tasks display as agent_task with a truncated prompt preview.

/sh

Drop into an interactive shell session without leaving muxd. Commands are executed directly — no agent involvement.

/sh

TUI only.
Type exit or press Ctrl+D to return to the muxd prompt.
Supports smart command routing on Windows (PowerShell/cmd detection).
The shell inherits the current working directory and environment.

/attach

Queue local files to include with your next message. Useful for screenshots, logs, and other files you want the daemon-side agent to analyze.

/attach                 # open local file picker
/attach <file-path>     # queue a specific file
/attach list            # show queued files
/attach clear           # clear queue

TUI only.
Queued files are sent with the next submitted prompt.
Directories are not allowed.

/remote-files

Browse the daemon host filesystem and copy files to your local TUI machine.

/remote-files

TUI only.
Space toggles multi-select on files.
Enter opens directories or copies selected file(s) to your current local working directory.
Backspace goes to the parent directory.

/transfer

Bidirectional file transfer between your local TUI machine and the daemon host.

/transfer remote   # remote -> local (open remote picker)
/transfer local    # local -> remote (open local picker)

TUI only.
/transfer remote uses the same remote browser as /remote-files and supports multi-select copy.
/transfer local uploads a selected local file to the daemon host.

/nodes

List and select hub nodes. Only available when connected to a hub.

/nodes

TUI only.
Shows all registered nodes with status (online/offline), host, and port.
Select a node to switch your active connection to it.

/qr

Display a QR code in the terminal for connecting the muxd mobile app. The QR code encodes the host, port, and authentication token.

/qr

TUI only.
Requires the daemon to be running (embedded or standalone).
If bound to 0.0.0.0, the QR code uses your LAN IP so mobile devices on the same network can connect.
The auth token is printed below the QR code for manual entry.
Cloud users (AWS, GCP, etc.): The IPs shown are local/private IPs (e.g. 172.31.x.x). These are not reachable from the public internet. Use your instance's public IP instead — find it in your cloud console or with curl ifconfig.me. Connect manually with muxd --remote <public-ip>:4096 --token <token>.

/remember

Save, remove, or list project memory facts. Facts persist across sessions in .muxd/memory.json and are automatically injected into the agent's system prompt.

/remember <key> <value>          # save a fact
/remember --remove <key>         # remove a fact
/remember                        # list all facts

Examples

/remember auth JWT with refresh tokens
/remember database SQLite with WAL mode
/remember test_patterns Table-driven tests, use t.Helper()
/remember --remove auth

See Project Memory for full details.

/consult

Ask a different AI model for a second opinion. Sends a question to the model configured as model.consult and displays the response in a separate styled block.

/consult [question]

If [question] is provided, it is sent directly to the consult model.
If no question is given, muxd auto-builds one from the last exchange in the current session.
Requires model.consult to be configured: /config set model.consult <model> (e.g. openai/gpt-4o, anthropic/claude-haiku).
The consult model receives no conversation history and no tools — only the focused question.
The primary agent does not see the consult response.

/mode

Switch the agent's chat mode. Three modes control which tools are available.

/mode <code|ask|architect>

Mode	Description
`code`	Default. Full tools: read, edit, run commands
`ask`	Read-only. No file writes or bash. Good for Q&A
`architect`	Plan-only. No writes. Agent can explore and propose without changing anything

Run /mode with no argument to see the current mode.

TUI only.

/tools

Show and toggle agent tools.

/tools

Opens a picker to enable/disable individual tools. Supports tool profiles: safe, coder (default), research.

/lsp

Show LSP server status and supported languages. Lists which language servers are installed and currently active.

/lsp

TUI only.
Install language servers to enable code intelligence for each language. See the LSP tool reference for install commands.

/refresh

Reload current session messages from the database.

/refresh

TUI only.

/exit

Quit muxd.

/exit

/quit

Alias for /exit.

/quit