@enatlas/sidecar

EnAtlas Sidecar — A local AI spend monitoring & optimization proxy.

Drop it in front of your OpenAI-compatible provider. It monitors every request, caches duplicates, routes between models, and reports telemetry to your EnAtlas dashboard — all without touching your prompts or API keys.

Quick Start

# 1. Set up your config interactively
npx @enatlas/sidecar init

# 2. Start the proxy
npx @enatlas/sidecar start

The init wizard will ask for your AI provider URL, API key, and your EnAtlas workspace credentials (found in your dashboard).

How It Works

The sidecar runs on http://localhost:4100 by default. Point your application's baseURL to the sidecar instead of your provider:

import OpenAI from 'openai';

const client = new OpenAI({
  baseURL: 'http://localhost:4100/v1', // ← EnAtlas sidecar
  apiKey: process.env.OPENAI_API_KEY,
});

// Use as normal — the sidecar proxies to your real provider
const res = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Hello' }],
});

Features

• Zero-code monitoring — Latency, tokens, cost, errors tracked automatically
• Exact-match caching — Identical prompts served from memory in <1ms
• Request coalescing — Concurrent duplicate requests collapsed into one upstream call
• Model routing — Downgrade models based on dashboard policies
• Resilience routing — Automatic failover to backup provider on 429/5xx
• Async telemetry — Never blocks your API calls

Configuration

The init command creates a .enatlas.env file. You can also set environment variables directly:

| Variable | Description | Default | |----------|-------------|---------| | UPSTREAM_BASE_URL | Your AI provider URL | — | | UPSTREAM_API_KEY | Your provider API key (stays local) | — | | CLOUD_INGEST_URL | EnAtlas telemetry endpoint | — | | WORKSPACE_ID | Your workspace ID | — | | WORKSPACE_INGEST_KEY | Your ingest key | — | | INTEGRATION_NAME | Provider name | openai | | PORT | Sidecar port | 4100 | | HOST | Sidecar host | 0.0.0.0 |

Trust Model

• Your API keys never leave your machine.
• Prompts and completions are never sent to EnAtlas cloud.
• Only metadata (model, tokens, latency, cost) is sent as telemetry.
• Telemetry is async and never blocks your API calls.

License

MIT