AI Provider types, registry, and configuration

[anurag629]

Phase 1 — Foundation (1/4)

Define the core type system and provider registry for the multi-provider AI integration.

Scope

src/lib/ai/types.ts

• AIProvider enum: openai, anthropic, google, groq, openrouter
• AIModel interface: { id, name, provider, maxTokens, description }
• AIProviderConfig interface: { id, name, baseUrl, models[], headerBuilder(apiKey), validateEndpoint }
• AIGenerateRequest interface: { provider, apiKey, model, prompt, maxTokens?, temperature? }
• AIGenerateResponse interface: { content, model, usage? }
• AIValidationResult interface: { valid, error?, provider }

src/lib/ai/providers.ts

• Provider registry map with config for each provider:
  • OpenAI: base URL, auth header format, available models (gpt-4o, gpt-4o-mini)
  • Anthropic: base URL, x-api-key + anthropic-version headers, models (claude-sonnet-4-6, claude-haiku-4-5)
  • Google: generativelanguage.googleapis.com, key-as-param, models (gemini-2.0-flash, gemini-2.5-pro)
  • Groq: OpenAI-compatible base URL, models (llama-3.3-70b, mixtral-8x7b)
  • OpenRouter: base URL, models (flexible — let users pick any)
• Helper functions: getProvider(id), getModelsForProvider(id), getDefaultModel(providerId)
• Each provider config includes the request body formatter (OpenAI-style vs Anthropic-style vs Google-style)

Acceptance Criteria

• All 5 providers defined with at least 2 models each
• Type exports are clean and well-documented
• Provider configs include correct API base URLs, header builders, and body formatters
• No runtime dependencies added — pure TypeScript definitions

Technical Notes

• OpenAI, Groq, and OpenRouter all use the OpenAI-compatible chat completions format
• Anthropic uses its own messages format (/v1/messages)
• Google uses the Gemini format (/v1beta/models/{model}:generateContent)
• Keep the registry extensible — adding a new provider should be one object addition

[anurag629]

Completed in dev branch.