Multi-Agent AI Orchestration Platform for Code Generation & Review
A production-ready platform for orchestrating AI agents to generate, review, and refine code through a structured workflow pipeline.
┌─────────────────────────────────────────────────────────────────┐
│ Frontend (React) │
│ Real-time Dashboard + Terminal │
└─────────────────────────┬───────────────────────────────────────┘
│ REST / SSE / WebSocket
┌─────────────────────────▼───────────────────────────────────────┐
│ API Gateway (Go) │
│ High-performance routing • JWT Auth • Rate Limiting │
└─────────────────────────┬───────────────────────────────────────┘
│ HTTP
┌─────────────────────────▼───────────────────────────────────────┐
│ Worker Service (Python) │
│ │
│ SUPERVISORY ┌──────────┐ ┌──────────┐ │
│ LAYER │ Architect│ <-> │ Planner │ iterate until │
│ └────┬─────┘ └────┬─────┘ consensus │
│ │ │ │
│ EXECUTION ┌─────▼────┐ ┌───────▼──┐ ┌──────────┐ │
│ LAYER │Orchestrat│->│ Coder │->│ Tester │ │
│ └──────────┘ └──────────┘ └────┬─────┘ │
│ │ │
│ INTEGRATION ┌────────────────────────────────▼──────────────┐ │
│ LAYER │ Integrator │ │
│ │ (consults Architect/Planner for merges) │ │
│ └───────────────────────────────────────────────┘ │
└─────────────────────────┬───────────────────────────────────────┘
│
┌─────────────────────────▼───────────────────────────────────────┐
│ Data Layer │
│ PostgreSQL (state) • Redis (cache) • S3 (artifacts) │
└─────────────────────────────────────────────────────────────────┘
- Architect Agent: High-level system design and technology decisions
- Planner Agent: Task breakdown and dependency ordering (iterates with Architect)
- Orchestrator Agent: Coordinates execution agents per task
- Coder Agent: Generates implementation based on approved specs
- Tester Agent: Reviews code quality, security, and test coverage
- Integrator Agent: Implements merges, consults supervisory agents
Unified provider abstraction supporting:
| Provider | Models | Auth Method |
|---|---|---|
| OpenRouter | GPT-5.2, Claude 4.5, Llama | API Key |
| Google Vertex AI | Gemini 3 Pro | Service Account |
| Amazon Bedrock | Claude, Titan | IAM Role |
| Microsoft Azure | GPT | API Key + Endpoint |
- SSE (Server-Sent Events): Live workflow progress updates
- WebSocket Terminal: Interactive PTY shell access
- Event-Driven Architecture: Pub/sub for agent coordination
- JWT Authentication with refresh token rotation
- OAuth 2.0 with Google and GitHub providers
- Multi-Factor Authentication (TOTP) with backup codes
- Session Management via Redis with device tracking
- Rate limiting with sliding window algorithm
- Connection pooling for database efficiency
- Graceful shutdown with request draining
| Layer | Technology | Rationale |
|---|---|---|
| Gateway | Go + Chi | Low latency, minimal memory footprint |
| Workers | Python + FastAPI | CrewAI ecosystem, async support |
| Database | PostgreSQL | ACID compliance, JSON support |
| Cache | Redis | Session storage, rate limiting |
| Frontend | React + TypeScript | Type safety, component reuse |
apps/
├── gateway/ # Go API Gateway
│ ├── cmd/server/ # Entry point
│ └── internal/
│ ├── auth/ # JWT middleware
│ ├── handlers/ # HTTP handlers
│ └── middleware/ # Rate limit, logging
│
├── api/ # Python Worker Service
│ ├── app/
│ │ ├── agents/ # CrewAI agent definitions
│ │ ├── workflows/ # Pipeline orchestration
│ │ ├── routers/ # FastAPI endpoints
│ │ └── llm_providers.py # Multi-cloud abstraction
│ └── alembic/ # Database migrations
│
└── console/ # React Frontend
└── app/
├── terminal/ # WebSocket terminal
└── components/ # Reusable UI components
- Go 1.22+
- Python 3.11+
- PostgreSQL 15+
- Node.js 20+
# Start database
cd apps/api && docker compose up -d db
# Run migrations
source .venv/bin/activate && alembic upgrade head
# Start Go gateway (port 8001)
cd apps/gateway && go run ./cmd/server
# Start Python workers (port 8002)
cd apps/api && uvicorn app.main:app --port 8002
# Start frontend (port 3000)
cd apps/console && npm run dev
⚠️ Security Requirements - The following environment variables MUST be configured for production:
# Required - Generate with: openssl rand -hex 32
export JWT_SECRET_KEY="your-secure-64-char-hex-key"
export OAUTH_ENCRYPTION_KEY="your-secure-64-char-hex-key"
# Required - Set environment
export KYROS_ENV="production"
# Required - Use SSL for database
export DATABASE_URL="postgres://user:pass@host:5432/db?sslmode=require"
# Required - Redis for sessions
export REDIS_URL="redis://host:6379"
# Required - Update CORS
export CORS_ALLOW_ORIGINS="https://your-domain.com"
⚠️ Database Warning: The defaultsslmode=disableis for development only. Production deployments MUST usesslmode=requireorsslmode=verify-full.
# Register
curl -X POST http://localhost:8001/auth/register \
-H "Content-Type: application/json" \
-d '{"username":"demo","email":"demo@example.com","password":"password123"}'
# Login
curl -X POST http://localhost:8001/auth/login \
-H "Content-Type: application/json" \
-d '{"email":"demo@example.com","password":"password123"}'# Create project
curl -X POST http://localhost:8001/projects \
-H "Authorization: Bearer $TOKEN" \
-d '{"name":"My App","description":"AI-generated application"}'
# Start generation
curl -X POST http://localhost:8001/projects/$ID/generate \
-H "Authorization: Bearer $TOKEN" \
-d '{"prompt":"Create a REST API for a todo application"}'| Challenge | Solution |
|---|---|
| CrewAI is Python-only | Keep agents in Python |
| API latency requirements | Go gateway for routing |
| Memory efficiency | Go uses ~10MB vs Python ~100MB |
| Development velocity | Python for rapid AI iteration |
- Vendor Independence: Avoid lock-in to single provider
- Cost Optimization: Route to cheapest provider per use case
- Reliability: Failover between providers
- Compliance: Use regional providers for data residency
# Health check
curl http://localhost:8001/health
# Provider status
curl http://localhost:8001/admin/providers
# Metrics (Prometheus format)
curl http://localhost:8001/metricsMIT