AtlasBurn
DocsPlatformIntegrationsOperationsReference

Proxy Quickstart

Zero install, zero latency, zero code changes

Swap one base URL and append one header. Your traffic now flows through the AtlasBurn Edge Proxy with telemetry capture and autonomous guardrail enforcement at <5ms.

1. Install — nothing

There is no SDK to install. The hosted proxy at proxy.atlasburn.com is a Cloudflare Worker that accepts traffic from any HTTP client in any language.

2. Get your AtlasBurn API key

Sign in to the AtlasBurn dashboard and copy your key from Settings → API Keys. Keys are prefixed abn_.

.env
bash
ATLASBURN_API_KEY=abn_xxxxxxxxxxxxxxxxxxxx
OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxx

3. Initialize the client

Point your existing AI client at the proxy. Two changes only: the base URL and the AtlasBurn-API-Key header. Pick your setup below — the provider's SDK, a raw HTTP call in any language, a framework, or a no-code tool. Every combination is one source of truth, so the snippet always matches what the proxy forwards.

Provider
Language
OpenAI · cURL
bash
curl https://proxy.atlasburn.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "AtlasBurn-API-Key: $ATLASBURN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "gpt-4o-mini",
  "messages": [
    {
      "role": "user",
      "content": "Hello from AtlasBurn"
    }
  ]
}'

Only two things change from a direct call: the base URL points at proxy.atlasburn.com/… and you add the AtlasBurn-API-Key header. Your provider key is forwarded untouched. Streaming works too — set stream: true and SSE frames pass through.

4. The event model

Every request through the proxy generates one telemetry event with this shape:

Telemetry event
json
{
  "ts": "2026-05-18T12:34:56.789Z",
  "org_id": "org_xxx",
  "provider": "openai",
  "model": "gpt-4o-mini",
  "endpoint": "/chat/completions",
  "status": 200,
  "tokens_in": 142,
  "tokens_out": 318,
  "cost_usd": 0.000421,
  "latency_ms": 612,
  "fingerprint": "sha256:b9c1...",
  "stream": false
}
  • Synchronous capture — non-streaming responses are parsed in-flight.
  • SSE pass-through — streamed frames forward untouched; the final usage frame is parsed and a single event is posted.
  • Fingerprint — deterministic hash of model + messages + tools, used by the Dumb Loop Detector.

5. Minimal end-to-end snippet

hello-proxy.ts
typescript
import OpenAI from "openai";

const openai = new OpenAI({
  apiKey: process.env.OPENAI_API_KEY,
  baseURL: "https://proxy.atlasburn.com/openai/v1",
  defaultHeaders: {
    "AtlasBurn-API-Key": process.env.ATLASBURN_API_KEY!,
  },
});

const res = await openai.chat.completions.create({
  model: "gpt-4o-mini",
  messages: [{ role: "user", content: "Say hi from the AtlasBurn proxy." }],
});

console.log(res.choices[0].message.content);
bash
npx tsx hello-proxy.ts

Verify in the dashboard

Open Forensic Ledger. The event appears within ~5 seconds with model, token counts, cost, and request fingerprint. Streamed responses without a final usage frame are flagged est/~est — see Streaming Usage.

Next steps