AI / LLM Integration

Connect multiple LLM providers and use them for text generation and multi-turn chat from a unified API.

Use case: Add AI-powered features to your application — content generation, Q&A, summarization — using whichever provider you have access to.

How It Works

1. Providers (OpenAI, Anthropic, Gemini, Ollama) are configured via the admin panel; API keys are stored in ai.config.json.
2. POST /api/ai/generate and POST /api/ai/chat proxy requests to the selected provider using native fetch.
3. Each call records token usage to ai.usage.json for persistent tracking.

Supported Providers

Provider ID	Label	Requires API Key	Default Models
openai	OpenAI	Yes	gpt-4o, gpt-4o-mini, gpt-4-turbo, gpt-3.5-turbo
anthropic	Anthropic	Yes	claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5-20251001
gemini	Google Gemini	Yes	gemini-2.0-flash, gemini-1.5-pro, gemini-1.5-flash
ollama	Ollama (Local)	No	llama3.2, mistral, codellama, phi3

Configuration

Providers are configured via:

1. Admin panel → AI Integration → Providers tab: Enable provider, paste API key, select default model.
2. Direct file edit: Edit ai.config.json in the project root.

Ollama requires a running local Ollama server. Default base URL: http://localhost:11434.

Token Usage Tracking

Every call to POST /api/ai/generate, POST /api/ai/chat, and POST /api/agents/:id/run records:

• Input token count
• Output token count
• Per-provider breakdown

Data is persisted to ai.usage.json and survives server restarts. Totals are visible in the admin dashboard.

AI Playground (Admin Panel)

• Generate mode: Single prompt with provider, model, system prompt, temperature, max tokens controls. Shows token count after each request.
• Chat mode: Multi-turn conversation with scrollable history and persistent context.
• Session totals: Cumulative token counts for the current browser session with a reset button.