Skip to main content

​
What Agent Chat is for

Agent Chat is the main execution surface. Prompts run against the active workspace folder with the currently selected model and agent profile.

​
Session controls

  • Agent selector: chooses behavior and policy profile.
  • Model selector: chooses inference model.
  • Clear UI: removes visible messages.
  • Delete context (memory): resets local chat context for the workspace.

​
Standard execution flow

  1. Select workspace folder.
  2. Select agent and model.
  3. Submit prompt.
  4. Review proposed plan.
  5. Approve or reject gated actions.
  6. Review outputs and file changes.

​
Prompting guidance for stable results

  • Give one concrete objective per prompt.
  • Include path boundaries when file edits are expected.
  • For destructive work, require confirmation in prompt instructions.
Example: 
Refactor only files in src/components/settings. Show a plan first and wait for approval before edits.

​
Keyboard behavior

  • Enter: send prompt
  • Shift + Enter: insert newline