CTA AI Nexus Router — Documentation

Welcome to the documentation for the CTA AI Nexus Router — a high-performance, OpenAI-compatible proxy that slashes LLM API costs through intelligent routing, context compression, and semantic caching.

Release Notes

v1.0.0

• Dual Build System: Now ships with both a lightweight CLI executable and a full GUI desktop application (Wails).
• Advanced Logging: Real-time memory log buffer in the UI, plus automatic background physical logs.txt writer inside the database folder.
• Enhanced Playground: Smooth auto-scroll behavior for long streams and quick "Delete Chat" session management.
• CORS Compatibility: Proper preflight OPTIONS support on proxy endpoints, enabling in-browser web UI fetching directly.
• Fixed image processing to support vision-capable models (e.g., gpt-4o, claude-3.5-sonnet).

Documentation Index

Getting Started

• Quick Start Guide — Install, run, configure, and start routing requests in minutes
• CLI Guide — How to run the lightweight headless router via command line

Architecture

• System Architecture — Directory structure, request flow, API endpoints, and data persistence
• Core Technologies — Deep dive into the compression pipeline, Lagrangian solver, discovery engine, and database schema

Routing

• Smart Routing — How complexity scoring, Lagrangian optimization, and the cascade fallback work

Providers

• Providers Guide — How provider resolution works, multi-key load balancing, payload normalization, and adding new providers

Compression

• Compression Engines — TOON extraction, Vision compression, the chunker, and telemetry

Quick Overview

Client (Cursor / IDE / API call)
    │
    │  POST http://localhost:20128/v1/chat/completions
    │  { "model": "cta-ai-nexus", "messages": [...] }
    ▼
CTA AI Nexus Router (Go, port 20128)
    │
    ├─ Intent Analysis → Complexity Score (0.0 – 1.0)
    ├─ Lagrangian Solver → Optimal Provider Cascade
    ├─ TOON Extraction → JSON compression via Groq
    ├─ Vision Compression → Image rendering via gg canvas
    │
    ▼
Upstream Provider (Groq / Mistral / Anthropic / Gemini / ...)
    │
    ▼
Streamed Response → Client

Port Reference