max tui opens an interactive terminal interface for chatting with Max. It connects to the running daemon at http://127.0.0.1:7777 via SSE (Server-Sent Events) and streams responses in real time with markdown rendering.
The daemon must be running before you open the TUI. Start it in a separate terminal with max start.
How it works
The TUI establishes a persistent SSE connection to the daemon’s /stream endpoint. When you type a message and press Enter:
- The message is POSTed to
/message on the daemon.
- The daemon enqueues it to the orchestrator.
- Response tokens stream back over the SSE connection and are rendered inline.
Proactive messages (e.g. worker completions triggered by background tasks) arrive over the same SSE stream and are printed with a MAX label whenever the TUI is open.
Connection
By default the TUI connects to http://127.0.0.1:7777. Override this with:
MAX_API_URL=http://remote-host:7777 max tui
~/.max/api-token and sent as an Authorization header on every request.
If the daemon is not reachable, the TUI retries every 5 seconds:
✗ cannot connect to daemon
start with: max start
Slash commands
Type a slash command at the prompt to control the daemon or the TUI itself.
| Command | Description |
|---|
/model [name] | Show the current model, or switch to a named model |
/auto | Toggle automatic model routing on or off |
/memory | List all stored memories |
/skills | List installed skills; enter a number to uninstall a local skill |
/workers | List active worker sessions with their status and working directory |
/copy | Copy the last response to the system clipboard |
/status | Print the daemon’s health check JSON |
/restart | Restart the daemon process in place |
/clear | Clear the terminal screen |
/help | Print the command reference inside the TUI |
/cancel | Cancel the current in-flight request |
/quit or /exit | Exit the TUI |
Switching models
/model
model: claude-sonnet-4.6
/model gpt-4.1
model: claude-sonnet-4.6 → gpt-4.1
Keyboard shortcuts
| Key | Action |
|---|
| Escape | Cancel the currently running response |
| Up / Down | Navigate input history |
| Ctrl+C | Exit the TUI |
Escape sends a /cancel request to the daemon and stops both streaming and the thinking indicator without exiting the TUI.
Real-time streaming
Responses stream token-by-token as they arrive. The TUI renders markdown to ANSI escape codes in real time:
- Headings (
#, ##, ###) are styled with bold/colour.
- Bullet and numbered lists are formatted with indent markers.
- Code blocks are surrounded by a dim
│ border.
- Inline
code, bold, and strikethrough are rendered with ANSI formatting.
A Thinking... spinner is shown while the daemon processes the request before the first token arrives.
History persistence
Every message you send (non-command lines) is appended to ~/.max/tui_history. The history file is capped at 1,000 entries and trimmed automatically on exit.
History is loaded into readline on startup so Up/Down navigation works across sessions.
Debug logging
Set MAX_TUI_DEBUG=1 to write a structured event log to ~/.max/tui-debug.log:
The log captures SSE lifecycle events, stream chunk details, and input rendering events as newline-delimited JSON.