IRENX

PRIME AI · OMNIROUTE · EDGE INTELLIGENCE

A self-hosted AI intelligence gateway built for precision, control, and elegant simplicity.

IRENX is intentionally simple on the surface and disciplined underneath. One gateway. One routing layer. Controlled integrations. Fail-fast CI. Regression gates before main.

✦ What is IRENX?

IRENX PRIME AI is a self-hosted AI gateway and intelligence terminal combining:

• OmniRoute Core Router V2 — task-aware provider routing, scoring, budget guards, timeout and circuit-breaker logic.
• OpenAI-compatible gateway — designed for OmniCopilot and compatible clients.
• Dify bridge — server-side workflows and chat integration.
• Live market layer — normalized market data endpoints for supported symbols.
• Cloudflare Rust/WASM canary — isolated edge runtime, promoted only after verification.
• Fail-fast CI + regression gates — broken changes stop before they reach main.
• Self-hosted deployment — Bun + Docker + Caddy, with no dependency on Vercel.

Design language

Black. Gold. Quiet. Precise.

The repository intentionally favors a premium, minimal visual identity instead of a crowded dashboard aesthetic.

◈ Architecture

Client
  │
  ▼
Public API / OpenAI-compatible Gateway
  │
  ▼
IRENX Core Router V2
  │
  ▼
OmniRoute
  │
  ├── GPT
  ├── Claude
  ├── Gemini
  ├── Qwen
  ├── DeepSeek
  └── other providers

Optional integrations
  ├── Dify
  ├── MCP
  └── Rust/WASM edge canary

Runtime authority

The TypeScript/Bun gateway remains the production authority. The Rust/WASM worker is a controlled canary, not an automatic replacement.

Promotion requires the relevant quality and runtime checks to pass.

◇ Repository structure

IRENX-ai/
├── index.html                 # PRIME terminal UI
├── api/                       # public API + AI gateway
│   └── v1/                    # OpenAI-compatible surface
├── src/omniroute/             # Core routing intelligence
├── mcp/                       # MCP integration surface
├── worker/                    # Cloudflare TypeScript worker
├── rust/                      # Rust/WASM edge canary
├── docs/                      # architecture + operations
│   └── assets/                # premium repository showcase graphics
├── .github/workflows/         # CI / governance / regression gates
├── Dockerfile                 # self-hosted image
├── docker-compose.yml         # production compose stack
├── Caddyfile                  # HTTPS reverse proxy
├── wrangler.toml              # Cloudflare configuration
└── deploy.sh                  # deployment + health checks

⚡ Fail-fast CI & regression gates

IRENX treats CI as a merge boundary, not a notification system.

PR / PUSH
   │
   ├─ source / governance validation
   ├─ formatting
   ├─ lint / Clippy
   ├─ regression tests
   ├─ WASM build
   ├─ worker-build validation
   └─ Wrangler dry-run
          │
          ▼
       ALL PASS
          │
          ▼
      eligible for main

The principle is strict:

If a required check fails, downstream verification stops and the change is not production-ready.

This prevents a formatting or compilation regression from being hidden behind later checks.

◎ API surface

Supported market symbols currently include XAUUSD, EURUSD, GBPUSD, USDJPY, and NAS100.

⌁ OmniRoute Core Router V2

IRENX adds an application-level policy layer in front of OmniRoute without duplicating OmniRoute's provider registry.

Resilience

• Circuit breaker with cooldown and recovery probing.
• Request timeout.
• Provider fallback delegated to OmniRoute.
• Observability around routing decisions.

Budget / quota guard

IRENX_AI_BUDGET_USD=2
IRENX_AI_MAX_REQUESTS=0
IRENX_AI_MAX_TOKENS=0
IRENX_INPUT_USD_PER_1M=3
IRENX_OUTPUT_USD_PER_1M=15
IRENX_EST_OUTPUT_TOKENS=1200

0 means unlimited for request/token quotas.

◇ OmniCopilot

VS Code / Copilot Chat
        │
        ▼
   OmniCopilot
        │
        ▼
https://ai.irenx.com/api/v1
        │
        ▼
 IRENX Core Router
        │
        ▼
    OmniRoute

By default, IRENX_COPILOT_RESPECT_MODEL=0, allowing IRENX to classify the workload and select a route family. Set it to 1 to respect the model selected by the client.

◇ Dify

IRENX provides a server-side Dify bridge:

• GET /api/dify/health
• POST /api/dify/workflows/run
• POST /api/dify/chat-messages

Dify remains an external/self-hosted application engine.

Deployment

IRENX is designed for a normal Linux VPS/server using Docker and Caddy.

Requirements

• Linux VPS/server
• Docker Engine
• Docker Compose plugin
• Ports 80/tcp and 443/tcp
• DNS ai.irenx.com → server IP

Environment

cp .env.example .env

At minimum:

OMNIROUTE_BASE_URL=http://127.0.0.1:20128
OMNIROUTE_API_KEY=...
TWELVEDATA_API_KEY=...

For Dify:

DIFY_BASE_URL=http://dify-api:5001
DIFY_API_KEY=...

Never commit .env, credentials, API keys, or provider secrets.

Start

bash deploy.sh

or:

docker compose up -d --build

Verify

curl -fsS https://ai.irenx.com/api/health
curl -fsS https://ai.irenx.com/api/ai/health

Security posture

IRENX follows a controlled architecture boundary:

• Credentials stay server-side.
• Public API access is separated from provider credentials.
• Rust/WASM is treated as a canary until verified.
• CI is designed to fail fast on regressions.
• Production deployment is self-hosted and explicit.

✦ Philosophy

LESS NOISE.
MORE SIGNAL.

SIMPLE INTERFACE.
SERIOUS ENGINEERING.

FAST FAILURE.
SAFE PROMOTION.

ONE GATEWAY.
FULL CONTROL.

IRENX PRIME AI

Precision over noise. Control over complexity.