Settings
The Settings page lets you configure the orchestration framework.
Oversight
Control how much autonomy agents have. Three tiers:
- Supervised — Agents ask before most actions. Best for learning or critical codebases.
- Balanced (default) — Routine work flows, structural changes need approval.
- Autonomous — Agents work independently. You monitor results.
Each tier injects behavioral instructions into agent system prompts. You can also add custom instructions in natural language (e.g., "Always run tests before committing").
Oversight is per-project with a global default. New projects inherit the global setting.
Concurrency
- Max Concurrent Agents — Maximum number of agents that can run simultaneously (default: 10)
- Auto-Restart — Automatically restart crashed agents (default: enabled)
- Max Restarts — Maximum restart attempts before giving up
Model Defaults
Override the default model for any role. Changes apply to newly created agents.
Custom Roles
Register new roles with:
- ID — Unique identifier (kebab-case)
- Name — Display name
- Icon — Emoji icon
- Color — Hex color for UI elements
- System Prompt — Instructions for the agent
- Default Model — AI model to use
Tool Permissions
Agents request tool permissions (file writes, shell commands) during operation. Permission timeout behavior depends on the agent's mode:
- Autopilot ON (lead-spawned or user-enabled): tool calls are auto-approved immediately
- Autopilot OFF (manually spawned): tool calls are auto-denied after 60 seconds if the user hasn't responded
Data Management
Monitor and manage the SQLite database that stores all session data.
- Database Statistics — File size, total records, oldest session date, per-table breakdown
- Purge Old Data — Select a retention period and preview what will be deleted before confirming
- Safety — Only completed sessions can be purged. Active sessions are protected. All deletions are transactional.