AtlasBurn
DocsPlatformIntegrationsOperationsReference

Getting Started

What is AtlasBurn?

AtlasBurn is an institutional AI capital risk engine. It provides forensic burn modeling and stochastic survival forecasting for teams running AI workloads at scale.

It is not a provider dashboard. It is not a routing layer. It is an economic forensics platform that operates at the SDK level across multiple providers.

The Problem

AI capital leakage happens fast and silently. Infinite agent loops, recursive retry cascades, and token surges bypass traditional provider billing alerts because those alerts are latency-heavy — often delayed 24–48 hours.

By the time a provider dashboard shows the spike, the budget is already gone.

Who It's For

  • Founders — protecting runway from silent AI cost blowouts
  • CFOs — forensic-grade capital accountability and forecasting
  • AI Infrastructure Engineers — real-time guardrails and telemetry at the SDK level

Core Value Proposition

Deterministic capital protection via the Auto Kill protocol and stochastic risk quantification using Monte Carlo simulations.

How AtlasBurn Is Different

ApproachFocusLimitation
Provider DashboardsUsage reporting24–48h latency, single-provider view
Routing ToolsPerformance optimizationNo economic forensics or risk modeling
AtlasBurnEconomic forensics + systemic riskSDK-level, multi-provider, real-time

What This Means in Practice

AtlasBurn catches the $400 agent loop before it becomes a $4,000 invoice. It gives you a forensic ledger entry for every single LLM call, a real-time burn rate, and deterministic budget guardrails that auto-kill runaway spend — all without touching your application logic.

Supported Providers

SDK auto-captured: OpenAI, Anthropic, Google Gemini, Google Vertex AI.
Captured via the Edge Proxy: the above plus OpenRouter (proxy routes: /openai/, /anthropic/, /gemini/, /openrouter/).
Priced but not yet auto-captured: Meta/Llama, Mistral, DeepSeek, xAI, Cohere — use the proxy or post events to /api/ingest.
Not yet supported: Azure OpenAI, AWS Bedrock, Groq, Together AI.

See the full Provider Matrix.

Platform Components

  • AtlasBurn SDK — @atlasburn/sdk v1.6.1, Apache-2.0. JavaScript / TypeScript auto-detect for Node, Edge, and the browser.
  • AtlasBurn Edge Proxy — a Cloudflare Worker that requires zero code changes (just swap the base URL). Works with any language, supports SSE streaming natively, and enforces guardrails at the edge.
  • AtlasBurn Cloud — the ingestion server, normalization engine, and forensic ledger.
  • AtlasBurn Dashboard — observability, reconciliation, and manual override of guardrail kill flags.

Supported Runtimes

  • Node.js
  • Vercel Edge
  • Browser (Chromium / WebKit)

System maturity: Stable

Core SDK, auto-detect, forensic ledger, and guardrails are production-ready.

Next Steps