Side Chat

Ask quick contextual questions without changing the main conversation

Side Chat lets you ask short questions about the current conversation without adding them to the main thread.

Use it when you need a quick clarification, summary, or next-step check but do not want to steer the main task.

What Side Chat is for

Good uses:

  • "What did we decide?"
  • "Summarize the last answer."
  • "What should I check next?"
  • "Explain this recommendation in plain language."

Avoid Side Chat when you want Claw to change files, run tools, continue the main plan, or create durable task state. Put those instructions in the main chat.

How it works

Side Chat uses:

  • The current main conversation as background context
  • Side Chat history for follow-up continuity
  • The current workspace and directory context
  • Working Brief and attached memory context
  • Workspace profile files when enabled
  • The selected working style
  • The Side Replies (/btw) model assignment when configured

Side Chat replies are brief by design. They should not assume the main task has changed.

Separation from the main thread

Side Chat is intentionally separate.

  • Side questions are not added as normal main-thread turns
  • Side Chat follow-ups stay in the side panel
  • The main task is not redirected by a side question
  • You can clear Side Chat content for the active conversation

This makes Side Chat useful for quick reasoning checks during long or complex work.

Tool behavior

Side Chat is for explanation, not execution. Side replies do not use shell, web, or web-search tool access.

If the answer requires an action, move the request into the main chat or hand the chat to an agent.

Model assignment

Side Chat can use a dedicated model through Settings > Model Assignments with Side Replies (/btw).

Use a fast, inexpensive model for Side Replies when:

  • The main model is expensive
  • Side questions are frequent
  • You mainly need summaries or clarifications

If no Side Replies model is configured, Claw falls back to the active conversation model where possible.