C
Docs

How C3 Works

Follow the journey of a message from your phone to your AI agent and back.

The Journey of a Message

When you send a message in C3, here's exactly what happens — from your fingertip to the agent's response appearing on your screen.

1. You Send a Message

Type your request in the C3 app — "Fix the login bug" or "Add a dark mode toggle" — and tap send.

2. Instant Delivery

Your message travels instantly to the C3 backend running on your computer. This happens in real-time — no delays, no queuing.

3. The Agent Gets to Work

Your AI agent receives your message and starts thinking. It analyzes your request, understands the context, and plans its approach.

4. Actions & Permissions

When the agent needs to read files, run commands, or make changes, it asks for your permission first. You see exactly what it wants to do before approving.

5. Results Appear

As the agent works, you see everything in real-time — responses appear as they're ready, file changes are highlighted, and command outputs display live.

Real-Time Updates

You don't wait for the agent to finish before seeing results. As it works, you see each step as it happens — tool executions, file reads, and responses appear in real-time.

A Real Example

Let's walk through a concrete example: asking your AI agent to fix a bug in your code.

You

"The login form is submitting twice when I click the button. Can you fix it?"

Agent is thinking...

"I need to find the login form component and check how the submit handler is implemented. This is likely a missing event.preventDefault() or a useEffect cleanup issue..."

Permission Request

The agent wants to read src/components/LoginForm.tsx

✓ AllowDeny
Agent

"Found the issue! The submit button doesn't have type="button" and the form's onSubmit isn't preventing default behavior. I'll fix both issues..."

This back-and-forth continues until the agent completes the task. You're always in control — approving each step, seeing every change, and able to guide the process with follow-up messages.

The Permission Flow

C3 gives you fine-grained control over what the agent can do. Here's how different actions are handled:

Always Allowed

  • • Reading file contents
  • • Searching code
  • • Listing directories
  • • Explaining code

Requires Approval

  • • Editing files
  • • Creating new files
  • • Running terminal commands
  • • Deleting files

Permission Modes

You can customize how permissions work based on your comfort level:

DefaultAsk permission for every file change and command
Auto-EditAuto-approve file edits, still ask for commands
Plan ModeThe agent only suggests changes, doesn't execute anything
YOLO ModeAuto-approve everything (use with caution!)

Start with Default Mode

We recommend using Default mode when you're getting started. Once you're comfortable with how the agent works, you can switch to faster modes for trusted projects.

Session Management

A session is a conversation with your AI agent about a specific project. C3 handles sessions intelligently so you can pick up where you left off.

Start Fresh

Begin a new conversation with full context of your project

Resume Anytime

Come back hours or days later — the agent remembers everything

Rewind Changes

Made a mistake? Roll back to any previous checkpoint

What Gets Saved

  • Conversation history — All your messages and the agent's responses
  • File changes — Every edit the agent made, with the ability to revert
  • Context — The agent remembers what files it has seen and what you discussed
  • Checkpoints — Automatic snapshots at key moments for easy rollback

Session Timeout

Sessions stay active for 30 minutes after your last message. If you come back later, C3 will either resume your session (if it's still available) or start fresh while keeping your conversation history accessible.

Staying Informed

You don't need to stare at your phone waiting for the agent. C3 keeps you updated even when you're doing other things.

Push Notifications

Get notified when the agent needs your approval or finishes a task. Tap the notification to jump right back into the conversation.

Status Indicators

See at a glance whether the agent is thinking, waiting for approval, or ready for your next message. No guessing required.

Notification Options

Customize notifications per server profile:

  • All Messages — Get notified for every agent response
  • Needs Attention — Only notify when the agent needs your input (permission requests, errors)
  • Off — No notifications (check manually when ready)

Background Work

Even when you're not looking at the app, the agent keeps working. Start a big refactoring task, put your phone down, and come back when it's done — you'll see everything the agent did waiting for your review.