AI Settings
Configure models, auto-run, edit approval, rules, and MCP servers.
Agent behavior is controlled from Settings → Agents. Open settings from the activity bar gear icon, Window → Settings, or Ctrl+,.
Models
| Setting | Purpose |
|---|---|
| Enabled models | Toggle which models appear in the chat picker |
| Default model | Pre-selected model for new threads |
Disable models you never use to simplify the dropdown.
Auto-run
Auto-run mode controls how the agent runs shell commands and other gated tools:
| Mode | Behavior |
|---|---|
| Ask every time | Prompt Run / Reject on each command |
| Auto (safe commands) | Default. Runs common safe commands without a prompt |
| Run everything | Runs commands except hard-blocked ones |
Protect destructive git (on by default) always asks before git reset, clean, restore, force-push, and similar.
See Reviewing Edits.
Edits
| Setting | Purpose |
|---|---|
| Require edit approval | Stage file edits for Accept / Skip before they are written |
| Auto-apply agent edits | Skip the Keep / Undo strip after edits land |
These are independent. Approval gates the write; auto-apply skips the review strip after the write.
Review
Adversarial review (on by default) runs critic models after Review mode finishes. Turn it off for a single pass.
Context
Max context lines limits how much surrounding code Shape sends when attaching a file. Higher values give the assistant more file context but use more credits.
Semantic codebase index builds on project open and after agent edits. It powers search_codebase. Re-index after large refactors.
Semantic embeddings (on by default when signed in) combine remote embeddings with keyword search. Off uses keyword search only.
Rules
The Rules field guides assistant behavior (style, conventions, files to avoid). Shape also loads .shape/rules.md from the project. These append to the built-in mode instructions for every message in the workspace.
MCP servers
Model Context Protocol servers extend the assistant with external tools. Configure servers in Settings → Agents → MCP.
Each server entry includes:
- Name and launch command
- Arguments and environment variables
- Enabled toggle
Restart or reconnect after changing MCP configuration.