@enatlas/sdk

EnAtlas SDK — In-process AI spend monitoring, caching & optimization for Node.js.

Wrap your OpenAI client to get automatic telemetry, exact-match caching, request coalescing, model routing, and resilience routing — all without a sidecar proxy.

Install

npm install @enatlas/sdk

Quick Start

import OpenAI from 'openai';
import { wrapOpenAI } from '@enatlas/sdk';

const openai = new OpenAI();

const { client, destroy } = wrapOpenAI(openai, {
  cloudIngestUrl: 'https://www.enatlas.com/api/ingest/events',
  workspaceIngestKey: 'en_your_key',
  workspaceId: 'your-workspace-id',
});

// Use client as normal — monitoring is automatic
const res = await client.chat.completions.create({
  model: 'gpt-4o',
  messages: [{ role: 'user', content: 'Hello' }],
});

// Clean up on shutdown
destroy();

Features

• Automatic telemetry — Latency, tokens, cost, errors tracked per request
• 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
• Per-instance isolation — Safe for multi-tenant and multi-client usage

Configuration

const { client, destroy } = wrapOpenAI(openai, {
  // Required — connects to your EnAtlas dashboard
  cloudIngestUrl: 'https://www.enatlas.com/api/ingest/events',
  workspaceIngestKey: 'en_...',
  workspaceId: 'uuid',

  // Optional
  integrationName: 'openai',    // Provider name for tagging
  maxCacheSize: 500,            // Max cache entries (FIFO eviction)
});

Trust Model

• Your API keys never leave your process.
• 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.

When to Use SDK vs Sidecar

| | SDK | Sidecar | |---|-----|---------| | Setup | npm install + wrap client | npx @enatlas/sidecar init | | Languages | Node.js only | Any (HTTP proxy) | | Latency | Zero network hops | 1 local hop | | Best for | Node.js apps wanting lowest overhead | Multi-language stacks, OpenClaw configs |

Both connect to the same dashboard and support the same optimization policies. You can switch between them without reintegration.

License

MIT