Terminal TUI

The terminal TUI is an interactive chat client built with Ink 5 and React 18 that connects to the Molf server over WebSocket, providing streaming responses, tool approval, session and workspace management, and slash commands directly in your terminal.

Starting the TUI

pnpm dev:client-tui

On first run, the TUI walks through two setup steps:

1. TLS fingerprint approval -- the server uses a self-signed TLS certificate by default. The TUI probes the certificate, displays its fingerprint, and asks you to confirm trust. The approved certificate is pinned to ~/.molf/known_certs/ for future connections.
2. Pairing -- if no auth token is available (via --token, MOLF_TOKEN, or saved credentials in ~/.molf/servers.json), the TUI runs an interactive pairing flow that exchanges a 6-digit code for an API key.

After setup, the Ink-based UI renders with a header, chat history, input area, and slash command autocomplete.

CLI Flags

Flag	Short	Env Var	Default	Description
--server-url	-s	MOLF_SERVER_URL	wss://127.0.0.1:7600	WebSocket server URL
--token	-t	MOLF_TOKEN	--	Auth token or API key
--worker-id	-w	MOLF_WORKER_ID	(auto-select first online worker)	Target worker UUID
--session-id	--	MOLF_SESSION_ID	(last session or new)	Resume an existing session by ID
--tls-ca	--	MOLF_TLS_CA	--	Path to a trusted CA certificate PEM file

If --worker-id is omitted, the TUI selects the first connected worker automatically. If --session-id is omitted, it loads the last session from the default workspace or creates a new one.

Connection

The TUI connects to wss://127.0.0.1:7600 by default (TLS). Authentication uses a Bearer token sent as a WebSocket header.

Token resolution order:

1. --token flag or MOLF_TOKEN env var
2. Saved API key from ~/.molf/servers.json (matched by server URL)
3. Interactive pairing flow (exchanges a 6-digit code for a new yk_-prefixed API key)

The WebSocket client reconnects automatically with exponential backoff (1 s initial, 30 s max, 2x multiplier, +/-25% jitter).

A warning is displayed when connecting to a remote server (not localhost / 127.0.0.1 / ::1) using a master token instead of a paired API key. The pinned certificate's expiry is also checked -- a warning appears if the certificate has expired or will expire within 30 days.

Slash Commands

Type / to enter command mode. An autocomplete popup appears with matching commands. Use Up/Down arrows to navigate and Tab to accept a completion.

Command	Aliases	Description
/clear	/new, /reset	Start a new session (the old session is preserved on disk)
/exit	/quit, /q	Exit the TUI
/help	/commands	Show all available commands
/sessions	/resume	Browse and switch sessions
/rename	--	Rename the current session (/rename <name>)
/worker	/workers, /w	List and switch between workers
/providers	/provider	Browse all LLM providers, add or update API keys at runtime
/model	/m	Two-step picker: first choose a provider, then choose from its available models (per-workspace)
/workspace	/ws	Browse and manage workspaces (/workspace new "name", /workspace rename "name")
/pair	--	Create a pairing code for a new device (/pair <device-name>)
/keys	--	List and remove stored LLM provider API keys
/editor	/edit, /e	Open $EDITOR to compose a message

Shell Shortcuts

Run shell commands directly on the worker, bypassing the LLM:

Prefix	Behavior
!<command>	Execute and save the output to session context (the LLM sees it on subsequent turns)
!!<command>	Execute fire-and-forget (output shown locally, not saved to context)

Examples:

!ls -la          # saved to context
!!git status     # fire-and-forget

The ! prefix requires the agent to be idle. If the agent is busy, use !! instead. Both dispatch via the agent.shellExec oRPC procedure.

Keyboard Shortcuts

Key	Action
Enter	Send message or execute slash command
Escape	Abort running agent if busy, or exit if idle
Ctrl+C	Exit immediately
Ctrl+L	Clear screen and start a new session
Ctrl+G	Open external editor to compose a message
Tab	Accept autocomplete suggestion
Up/Down	Navigate autocomplete, or scroll through input history

Tool Approval

When the LLM requests a tool call that requires user permission, the TUI displays an inline approval prompt showing the tool name and its arguments. The input bar is disabled until the approval is resolved.

Key	Action
Y	Approve this single tool call
A	Always approve matching tool+pattern calls (persisted to permissions.jsonc)
N	Deny -- opens a feedback text input where you can type an optional reason, then press Enter

When multiple approvals are pending, a [1/N] counter shows your position in the queue.

If the TUI disconnects and reconnects, any pending approval prompts are automatically replayed by the server.

See Tool Approval for how rules are evaluated.

Pickers

Several commands open full-screen interactive pickers:

• Workspace picker (/workspace) -- browse workspaces and their sessions, create or rename workspaces, select a session to switch to
• Session picker (/sessions) -- browse and switch between sessions
• Worker picker (/worker) -- list workers with their tool counts and online/offline status
• Provider picker (/providers) -- browse all 16 supported LLM providers; shows key status (env for environment variables, stored for runtime-added keys, or no key) and model count per provider; select a provider to add or update its API key, or to set the default model
• Model picker (/model) -- two-step: first select a provider (from providers that have a key), then select a model from that provider; the chosen model is set as the workspace model, or reset to the server default
• Key picker (/keys) -- list stored LLM provider API keys and remove them

All pickers support arrow key navigation and Escape to cancel.

Interface Layout

The TUI renders the following sections from top to bottom:

• Header -- connection status, worker name, workspace name, and keyboard shortcut hints
• Chat history -- user, assistant, and system messages
• Active tool calls -- tools currently being executed by the worker
• Subagent blocks -- progress of active subagent tasks
• Streaming response -- real-time LLM output as it arrives
• Status bar -- current agent status (idle, streaming, executing_tool)
• Cron notifications -- alerts when scheduled tasks fire in other sessions
• Tool approval prompt -- when a tool call needs user permission
• Input bar -- multi-line text area
• Autocomplete popup -- slash command completions

See Also

• Getting Started -- quick-start guide
• Configuration -- all CLI flags and env vars
• Telegram Bot -- alternative Telegram client
• Building a Custom Client -- using the oRPC API directly
• Tool Approval -- approval rules and permissions.jsonc