Balanced mode is the middle ground between full control and full automation. Chapeta classifies each tool action as safe or potentially dangerous. Safe actions (reading files, web searches, non-destructive commands) execute automatically. Dangerous actions (writing files, running commands that modify state) require your approval.
How It Works
- The AI determines it needs to use a tool.
- Chapeta classifies the action as safe or potentially dangerous.
- Safe actions execute immediately without a prompt.
- Dangerous actions show an approval prompt, just like Supervised mode.
- You can see a log of all auto-approved actions in the conversation.
Benefits
- Significantly faster than Supervised mode for common workflows.
- Read operations and lookups happen instantly.
- Destructive or state-changing operations still require approval.
- Good daily driver for users who understand the tool capabilities.
Tradeoffs
- Some actions run without your review. You trade visibility for speed.
- The safe/dangerous classification is rule-based, not perfect.
- May auto-approve actions you would prefer to review in specific contexts.
FAQ
More Safety Features
Supervised Mode
Approve every tool action before it runs. Zero surprises.
Autopilot Mode
Full automatic execution. No approval prompts. Maximum speed.
God Mode
All guardrails disabled. Full autonomy. You accept full responsibility.