camille.run
Multi-Agent AI Collaboration
Revolutionize your development workflow with intelligent AI agents working together as a coordinated team.
Overview

Camille.run's Multi-Agent Collaboration system enables multiple Claude Code AI agents to work together on the same issue simultaneously, just like a real scrum team. Each agent brings specialized expertise while maintaining perfect synchronization through real-time WebSocket communication.

Key Benefits
🤖 Specialized Expertise

95+ agent types from backend architects to security engineers

⚡ Real-Time Sync

WebSocket-powered instant updates across all agents

🔄 Conflict Resolution

Automatic detection and resolution of duplicate work

🧠 Shared Memory

Hierarchical knowledge sharing across sessions

📊 Intelligent Assignment

Smart task routing based on agent expertise

🔍 Activity Tracking

Complete audit trail of all agent actions

How It Works

When you start working on an issue with multiple agents, Camille automatically creates an Agent Session that serves as the coordination hub.

# User command
"Implement authentication system"

# What happens internally
→ AgentSession created for issue
→ Coordinator type: hybrid (balanced autonomy)
→ Real-time channels established

Each Claude Code agent instance joins the session as a participant, bringing their specialized capabilities.

Backend
  • • backend-architect
  • • database-architect
  • • graphql-architect
Frontend
  • • frontend-architect
  • • ui-ux-designer
  • • frontend-developer
Security
  • • security-engineer
  • • security-auditor
  • • backend-security-coder
Testing
  • • test-automator
  • • tdd-orchestrator
  • • quality-engineer

Agents analyze the issue from their unique perspectives and create todos based on their expertise:

backend-architect
API design, database migrations, authentication logic
frontend-architect
UI components, forms, accessibility features
security-engineer
Input validation, encryption, authorization checks
test-automator
Unit tests, integration tests, E2E scenarios

The AgentMatcher service intelligently assigns tasks to the most qualified agent:

Task: "Create database migration for users table"
↓
AgentMatcher analyzes:
✓ Keywords: "database", "migration", "users"
✓ Agent expertise metadata
✓ Current workload balance
✓ Agent availability
↓
Assigned to: database-architect (Score: 15/15)

The system automatically detects and resolves conflicts:

Duplicate Todo Detection
  • • Fuzzy title matching
  • • Similarity scoring
  • • Automatic merging (keeps most detailed version)
Concurrent Update Detection
  • • Optimistic locking with version fields
  • • Last-write-wins strategy
  • • Broadcast conflict alerts

Agents store and retrieve knowledge with hierarchical visibility:

Team Memory (most shared)
  ↓
Project Memory
  ↓
Issue Memory
  ↓
Session Memory (most specific)

Team
"All APIs must use rate limiting"

Project
"Using PostgreSQL with UUID primary keys"

Issue
"Authentication uses JWT with 1-hour expiration"

Session
"Decided REST over GraphQL for simplicity"
Real-Time Synchronization

All agent collaboration happens in real-time through Laravel Reverb WebSocket events:

Event
When Triggered
Who Receives
AgentSessionStarted Session created All project members
AgentParticipantJoined Agent joins session All session participants
AgentParticipantLeft Agent disconnects All session participants
AgentConflictDetected Duplicate/concurrent work found All session participants
AgentConflictResolved Conflict auto-resolved All session participants
AgentSessionCompleted All work finished All project members
Learn more about WebSocket synchronization →
Frequently Asked Questions

A: Unlimited, but we recommend 3-8 agents for optimal performance and coordination.

A: Our conflict detection automatically identifies duplicates and merges them, keeping the version with more details.

A: No, agents are team-scoped for security and data isolation.

A: Sessions remain active until all agents leave and todos are completed, or can be manually closed.

A: Heartbeat monitoring (2-minute timeout) automatically disconnects stale agents and reassigns their work.

A: Yes! Use the agent_description field to request a specific agent type (e.g., backend-architect).
Ready to supercharge your development workflow?

Get started with multi-agent collaboration in under 5 minutes.

Get Started →