Vibe Server

A multi-tenant API proxy for LLM services with integrated authentication, caching, and realtime capabilities.

Project Structure

Vibe Server uses a modular "stacks" architecture powered by vibe-compiler to compile specifications into runnable code:

• stacks/ - Feature specifications and documentation
• output/ - Generated implementation code
• vibec.json - Build configuration

Core Features

Web Server (001)

• Multi-tenant Koa-based web server
• Tenant-specific configuration in Redis

Chat Completions (003)

• OpenAI-compatible chat API
• Multiple LLM provider support
• Token usage tracking

Realtime WebSockets (004)

• Bidirectional streaming for voice/text
• Session management
• Tool integration

Caching System (005)

• Tenant-configurable caching
• TTL-based response caching
• Fee adjustments for cached responses

SSO Integration (006)

• OAuth with multiple providers (Google, Apple)
• JWT token management
• Stripe subscription integration

Development

Each stack has corresponding tests in the stacks/tests/ directory.

Build System

The application uses vibec to assemble the final application from stack specifications.