Moltis
Features

Slash Commands

Slash commands are available in the web UI chat input, on all messaging channels (Telegram, Discord, Slack, Matrix, etc.), and where noted, via the CLI.

Type / in the chat input to see the autocomplete popup.

Session Management

CommandDescription
/newStart a new session
/clearClear session history
/compactSummarize conversation to save tokens
/contextShow session context and project info
/sessionsList and switch sessions (channels only)
/attachAttach an existing session to this channel (channels only)
/fork [label]Fork the current session into a new branch

/fork

Creates an independent copy of the current conversation. The new session inherits the parent’s model, project, mode, and agent. Messages up to the current point are copied.

/fork experiment-a

Available in web UI, all channels, and via the sessions.fork RPC. See Session Branching for details.

Control

CommandDescription
/agent [N]Switch session agent
/mode [N|name|none]Switch session mode
/model [N]Switch provider/model
/sandbox [on|off|image N]Toggle sandbox and choose image
/sh [on|off]Enter command mode (passthrough to shell)
/stopAbort the current running agent
/peekShow current thinking/tool status
/update [version]Update moltis (owner-only)

Quick Actions

CommandDescription
/btw <question>Quick side question (no tools, not persisted)
/fast [on|off|status]Toggle fast/priority mode
/insights [days]Show usage analytics (tokens, providers)
/steer <text>Inject guidance into the current agent run
/queue <message>Queue a message for the next agent turn
/rollback [N|diff N]List or restore file checkpoints

/btw

Ask a quick side question without tools and without persisting the exchange to session history. Uses the session’s current model and recent context (last 20 messages) as read-only background.

/btw what's the default port for PostgreSQL?

The response appears inline and is discarded after display.

/fast

Toggle fast/priority mode for the current session. When enabled, uses provider-specific priority processing where supported (Anthropic prompt caching priority, OpenAI priority processing).

/fast          # toggle
/fast on       # enable
/fast off      # disable
/fast status   # check current state

Session-scoped — does not persist across gateway restarts.

/insights

Show usage analytics from the metrics store. Displays LLM completions, token counts (input/output), errors, tool executions, and per-provider breakdowns.

/insights       # last 30 days (default)
/insights 7     # last 7 days
/insights 90    # last 90 days

In the web UI, /insights renders a formatted markdown table inline. The same data is available as a dashboard in Monitoring > Insights tab, and via the REST API at GET /api/metrics/insights?days=N.

/steer

Inject guidance into an active agent run without interrupting it. The text is seen by the LLM on its next iteration (after the current tool call completes).

/steer use the staging API, not production
/steer focus on security issues only

Only works while an agent run is active. If no run is active, returns an error.

/queue

Queue a message for the next agent turn without interrupting the current one. When the active run finishes, the queued message is automatically submitted.

/queue now write tests for what you just built

If no run is active, the message is sent immediately.

/rollback

List and restore file checkpoints created by the automatic checkpointing system. Before every Write, Edit, or MultiEdit tool call, the original file is snapshotted.

/rollback           # list recent turns with file changes
/rollback 1         # restore all files from turn 1
/rollback diff 1    # preview which files were changed in turn 1

Checkpoints are grouped by turn (one user message = one turn). Restoring a turn reverts all files that were modified during that turn to their pre-turn state.

See Checkpoints for details on the automatic checkpointing system.

Approval Management

CommandDescription
/approvalsList pending exec approvals
/approve [N]Approve a pending exec request
/deny [N]Deny a pending exec request

Help

CommandDescription
/helpShow available commands (handled locally by each channel)