​

Architecture

​

System Overview

Client → api.modelbeam.ai → Job Queue → GPU Workers → Storage
                                 ↓
                          Status Updates
                    (Polling / Webhooks / WebSockets)

​

Components

​

API Gateway (api.modelbeam.ai)

• Handles authentication, rate limiting, and request validation
• Creates job records and dispatches to GPU workers
• Manages billing (balance checks, price calculation, deductions)
• Serves job status via polling endpoint

​

GPU Workers

• Serverless GPU instances running AI models
• Auto-scale based on demand
• Send status callbacks as jobs progress
• Upload results to object storage

​

Storage (storage.modelbeam.ai)

• Stores generated files (images, audio, video)
• Results available via direct download URLs
• Temporary storage with configurable retention

​

Real-time Layer (soketi.modelbeam.ai)

• Pusher-compatible WebSocket server
• Pushes real-time status updates to connected clients
• Private channels per client for security

​

Integration Points

​

Authentication Flow

1. User registers at modelbeam.ai and receives $5 free credits
2. User creates an API key from the dashboard
3. API key is sent as Bearer token in the Authorization header
4. API validates the key, checks rate limits, and processes the request

​

Job Flow

1. Client sends a POST request to a generation/analysis endpoint
2. API validates parameters, checks balance, calculates price
3. Balance is deducted and a job record is created (status: pending)
4. Job is dispatched to a GPU worker
5. API returns {"data": {"request_id": "UUID"}}
6. Worker processes the job, sends progress updates
7. On completion, results are uploaded to storage
8. Client receives results via polling, webhook, or WebSocket