[namanmukund]

🤖 Summary of Changes

This PR introduces the Universal AI Schema Copilot (BYO-LLM) into AutoGraphQL Studio (/studio). It allows developers to generate and refine production-grade GraphQL SDL schemas using natural language prompts, with native understanding of AutoGraphQL's custom AST directive system (@model, @tenantScoped, @ownerScoped, @history, @relation, @clamp, @validate, etc.).

🌟 Key Highlights & Capabilities

1. Universal Multi-Provider Architecture (BYO-LLM):
   • Google Cloud Vertex AI (Gemini): Native Google SDK authentication via Application Default Credentials (gcloud auth application-default login) — supports gemini-2.5-flash, gemini-2.0-flash, and gemini-1.5-pro with zero client-side token pasting.
   • Local & Offline Open-Source Models: First-class support for Ollama (http://localhost:11434/v1 with llama3.2, qwen3, etc.), vLLM, LocalAI, and LM Studio.
   • Commercial Cloud Providers: OpenAI (gpt-4o, o3-mini), Google Gemini API (gemini-2.0-flash), Anthropic Claude (claude-3-5-sonnet), Groq (llama-3.3-70b), and OpenRouter.
   • Custom Gateways: Configurable base URLs for private self-hosted LLM clusters.
2. 🔒 Zero-Server Storage (Privacy-First Security):
   • API keys and provider configurations are persisted exclusively in the browser's localStorage.
   • Keys are transmitted ephemerally via secure request headers (x-ai-provider, x-ai-key, x-ai-model, x-ai-base-url) and never written to disk, database, or server logs.
3. AutoGraphQL AST Directive Nomenclature Knowledge Base:
   • The AI engine generates schemas strictly adhering to AutoGraphQL syntax:
     • Database Backing: @model(database: postgres) or MongoDB default.
     • Multi-Tenancy & RLS: @tenantScoped(field: "tenantId", claim: "tenantId"), @ownerScoped.
     • Audit Trails: @history.
     • Relational Connectors: @relation(name: "...", direction: "OUT" | "IN" | "BOTH") and @relationalMeta.
     • Field Constraints: @clamp(min, max), @validate(regex), @defaultValue(value), @unique.
     • Field Sanitization & Transforms: @trim, @nameCase, @upperCase, @lowerCase, @slug.
     • Field Security: @readOnly, @writeOnly, @filterOff, @encrypted.
     • Custom Types & Enums: Generates valid enum Status { ... } alongside types.
4. 🔄 RAG Active Schema Awareness:
   • Scans existing .graphql files in schemas/ and dynamically injects active entities into the system prompt context so the AI can automatically link foreign keys and relationships to existing models (e.g. User, Category).
5. 🛡️ AST Validation, Local Normalization & Self-Healing:
   • Pre-normalizes local model output quirks (e.g. model Course $\rightarrow$ type Course).
   • Parses generated SDL with graphql.parse. If an AST syntax error is detected, an automated self-healing retry loop provides AST error feedback to correct the schema before presenting it to the user.
6. ⚡ Bi-Directional Visual Builder Sync:
   • ⚡ Sync to Visual Builder: Dynamically parses generated or edited SDL into the Visual Model Builder cards, fields, constraints, and custom enums.
   • Deploy Model ⚡: 1-click deployment compiles the schema straight into schemas/<Model>.graphql.

---