Safety & Ethics
ChatPharo includes guardrails to help users avoid accidental misconfiguration and to keep AI-assisted workflows safer.
Safety Advisor (confirmations and defaults)
The Safety Advisor can:
- propose recommended defaults (agent, tools, skills, caching, sandbox, etc.)
- show confirmations when changing high-impact settings, such as:
- modifying prompts
- disabling tools
- lowering iterations below 1
- changing message limits
- selecting cloud agents (external API warning)
External API warning
When you pick an agent that uses an external API (cloud), ChatPharo can show a warning dialog about sending data to an external provider. Verified developers can optionally dismiss this warning permanently.
Ethics settings
Ethics controls may include budgets and limits such as:
- token budget
- time budget
- message limit
- cognitive-load budget
Ethics settings are commonly restricted to verified developers and require enabling a “developer” mode or providing a developer ID.
Practical guidance
- Keep tools enabled for best results; disable only when diagnosing.
- Keep tool iterations bounded (e.g., 2–5).
- Prefer local agents if you handle sensitive code and do not want data to leave your machine.
- Configuration: configuration.md
- Sandbox: sandbox.md