- I already track my costs in my providers' billing dashboards.
- Billing dashboards show you what already happened. AtlasBurn helps you understand what's likely to happen next — and gives you guardrails to act on it before the next invoice.
- My burn is fine right now.
- That's great. But AI API costs tend to change faster than expected as usage scales. AtlasBurn helps you stay ahead of that curve so there are no surprises.
- This sounds like another monitoring tool.
- Monitoring shows you metrics. AtlasBurn adds forecasting and cost controls on top — so you're not just watching numbers, you're setting guardrails that protect your runway.
- Do you see my prompts or my users' data?
- No. AtlasBurn stores usage numbers only: model, tokens, cost, time and the feature label you choose. Prompts and outputs are never stored.
- What happens to my app if AtlasBurn goes down?
- With the SDK, nothing. It fails open: if it can't reach AtlasBurn, your AI calls go through as normal. With the edge proxy, requests pass through AtlasBurn's edge on Cloudflare, so they need it to be reachable. Either way, AtlasBurn only slows down or blocks traffic when your own guardrail settings say to.
- Can't I just build this myself?
- You could, but it takes meaningful engineering time to build cost tracking, forecasting, and alerting across multiple AI providers. AtlasBurn handles that so your team can focus on product.
- What is AtlasBurn?
- AtlasBurn is an AI API cost monitoring and guardrail tool. It tracks what each feature of your app spends across 13 LLM providers, forecasts your monthly AI bill, and can throttle or block a feature that gets stuck in a loop or crosses a budget, before the cost reaches your invoice.
- How does AtlasBurn stop a runaway AI agent loop?
- It watches each feature's recent calls. If the same request repeats several times within 30 seconds, or the context keeps growing until it has at least doubled, that feature is flagged as looping. On the Kill Switch tier it is first slowed to one call every 2 seconds; if the loop is still going 60 seconds later, or has cost more than your limit, it is blocked. Only the looping feature is affected; the rest of your app keeps running.
- Which AI providers does AtlasBurn work with?
- The SDK auto-detects OpenAI, Anthropic, Google Gemini, Google Vertex AI, Azure OpenAI, AWS Bedrock, OpenRouter, Cohere, Mistral, Groq, Together AI, DeepSeek and xAI. The Edge Proxy covers OpenAI, Anthropic, Gemini, OpenRouter, Mistral, Cohere, Groq, Together AI, DeepSeek and xAI from any language.
- Can I use AtlasBurn from Python or another language?
- Yes. The SDK is for JavaScript and TypeScript, but the Edge Proxy works with any HTTP client, including the official Python SDKs: point the base URL at proxy.atlasburn.com and add an AtlasBurn-API-Key header. You can also post token counts to the ingest API yourself.
- How is AtlasBurn different from LLM observability tools like Helicone or Langfuse?
- Observability tools log requests and responses so you can trace, debug and evaluate them. AtlasBurn is built for spend: it never stores prompts or outputs, attributes cost to each feature, forecasts the bill, and can slow down or stop a feature at runtime when it loops or breaks a budget. Many teams use an observability tool for debugging and AtlasBurn for cost control.
- How much does AtlasBurn cost?
- AtlasBurn is free during beta. After beta the Free plan stays free for up to 10M AI tokens a month; paid plans start at $19 a month for 50M tokens. Guardrails, the kill switch, unlimited read-only viewers and a full data export are included on every plan.
- Can I ask Claude or ChatGPT about my AI spend?
- Yes. AtlasBurn has an MCP server, listed in the Claude connector directory and in ChatGPT. Connect it and ask questions like “what did we spend this week, and why?”. In Claude Code you can also say “Integrate AtlasBurn” to set it up; you approve the new key in AtlasBurn, and it never appears in the chat.
%20translate(0,0)'/%3e%3c/svg%3e){kind=small}
