Advanced Settings

Permission Modes

Permission modes control how much autonomy the agent has during a session. You can set the mode when starting a new session or change it mid-session.

Available Modes

• Default — The agent asks permission before file edits, command execution, and other actions. Best for careful, supervised work.
• Plan — The agent creates a detailed plan first, then asks for your approval before implementing. See Plan Mode below.
• Accept Edits — The agent can edit files without asking, but still requests permission for commands and other actions.
• Bypass Permissions — The agent runs fully autonomously without any permission prompts.

Changing Modes Mid-Session

You can switch permission modes during an active session. C3 seamlessly applies the new mode while preserving your session history.

Plan Mode

Plan mode is a two-phase workflow where the agent first creates a plan, then implements it after your approval.

How It Works

1. Start a session in Plan mode (or switch to it mid-session)
2. The agent analyzes your request and produces a detailed implementation plan
3. You review the plan and approve, modify, or reject it
4. On approval, the agent implements the plan

Clear Context

When approving a plan, you can choose to clear context. This starts a fresh session with only the approved plan as context, discarding the conversation history. This is useful when the planning conversation was long and you want the agent to focus purely on execution.

Model Selection

Choose which model to use for your sessions. You can set a default in Settings and override it per session.

• Haiku — Fastest and most cost-effective. Great for quick questions, simple edits, and lightweight tasks.
• Sonnet — Balanced speed and capability. The default choice for most development work.
• Opus — Most capable model. Use for complex architecture decisions, large refactors, and tasks requiring deep reasoning.

Appearance & Haptics

Theme

Choose between Light, Dark, or System theme. The System option follows your device's appearance setting automatically.

Haptic Feedback

Toggle haptic feedback on or off. When enabled, you'll feel subtle vibrations on interactions like button presses and navigation actions.

MCP Servers

Configure Model Context Protocol (MCP) servers to extend the agent's capabilities with custom tools and data sources. MCP servers are configured per server profile.

Transport Types

• HTTP — Connects to an HTTP endpoint. Requires a URL.
• SSE — Server-Sent Events transport. Requires a URL.
• Stdio — Spawns a local process. Requires a command.

Configuration Options

Each MCP server can be configured with:

• Arguments — Command-line arguments (for Stdio type)
• Environment Variables — Key-value pairs passed to the server process
• HTTP Headers — Custom headers for HTTP/SSE connections

Slash Commands & Memory

Each server profile's Advanced section provides access to several configuration areas:

• Slash Commands — Define custom /command shortcuts that expand into longer prompts or workflows
• Agent Memory — Manage CLAUDE.md memory files that persist context across sessions
• Hooks — Configure shell commands that run in response to agent events
• Output Styles — Customize how the agent communicates (tone, verbosity, formatting)
• Plugins — Manage agent plugins from marketplaces or custom sources

Notifications

Notification preferences are configured per server profile in the Advanced section.

Notification Modes

• All Messages — Get notified for every message from the agent. Useful when you're multitasking and want to stay updated.
• Needs Attention — Only notify when the agent needs your input (permission requests, questions, errors). Reduces noise during long autonomous tasks.

You can also disable notifications entirely for specific profiles — useful for profiles used in local development where you're already watching the screen.

Server Profile Management

Server profiles store connection details for your C3 backend instances. You can create multiple profiles for different machines or environments.

Creating Profiles

Tap the + button to add a new profile. Each profile needs:

• Name — A friendly label (e.g., "Work Laptop", "Home Desktop")
• Server URL — The C3 backend URL
• Username & Password — Authentication credentials
• SSL/TLS — Enable for HTTPS connections

QR Code Scanner

The fastest way to add a profile: tap the QR scanner icon in the profile modal header and scan the QR code displayed by your C3 backend on startup. This fills in all connection details automatically.