AI Service

Product Factory provides a unified AI service through the worker-ai service. It acts as a gateway between your application and multiple AI providers, handling model routing, usage tracking, and credit deduction.

Architecture

Frontend (React Island)
  → Gateway (/ai/*)
    → worker-ai
      → Cloudflare AI Gateway / Direct Provider API
        → OpenAI / Anthropic / Google / etc.

All AI requests go through the gateway. The frontend never connects directly to AI providers.

Providers and Models

AI providers and models are managed through the admin console. Each provider has:

Field	Description
name	Provider identifier (e.g., “openai”, “anthropic”)
api_key	Encrypted API key
base_url	Provider API endpoint
enabled	Whether the provider is active

Models are registered per provider:

Field	Description
model_id	Model identifier (e.g., “gpt-4”, “claude-3-opus”)
provider_id	Parent provider
credit_cost	Credits per 1K tokens
enabled	Whether the model is available to users

Chat Completions

The primary AI endpoint is /ai/chat, which accepts messages and returns completions:

curl -X POST http://localhost:8787/ai/chat \
  -H "Content-Type: application/json" \
  -H "Cookie: better-auth.session_token=..." \
  -d '{
    "model": "gpt-4",
    "messages": [
      {"role": "user", "content": "Hello"}
    ]
  }'

The service handles:

1. Session verification (via gateway)
2. Credit preflight check
3. Model routing to the correct provider
4. Response streaming (when supported)
5. Usage tracking and credit deduction

Usage Tracking

Every AI request is logged with:

Field	Description
user_id	Requesting user
model_id	Model used
prompt_tokens	Input token count
completion_tokens	Output token count
credit_cost	Credits deducted
duration_ms	Request duration

API Endpoints

See the AI API Reference for the complete endpoint documentation.