AtlasBurn

AtlasBurn connects Claude to your real AI usage and cost data so you can ask where your LLM spend is going and catch problems before they become a large bill. What you can ask: - Total AI spend, token usage, and request volume over any window - Cost broken down by model and by product feature, so you can see what is actually driving the bill - Projected monthly bill, current daily burn, and budget runway - Guardrail status: whether budgets are enabled, the enforcement state (active, throttled, or suspended), any single feature paused after a runaway loop, configured daily and hourly limits, and recent breaches - Runtime signals such as retry rate, loop probability, and context growth, to spot an agent stuck in a retry loop or running away Set AtlasBurn up from Claude: ask "Set up AtlasBurn in this project." Claude requests a key, you approve it in your browser with a fresh sign-in, and the AtlasBurn CLI receives the new key and writes it to your project's env file. The key never passes through Claude or the chat, and your existing keys are never changed. AtlasBurn works across 13 providers, including OpenAI, Anthropic, Google Gemini, Vertex AI, Azure OpenAI, AWS Bedrock, Mistral, Groq, DeepSeek, xAI and OpenRouter. Usage data comes from the open-source AtlasBurn SDK (Node.js/TypeScript) or the AtlasBurn proxy for any other language. The connector is content-blind: it never sees your prompts or model outputs. The data tools are read-only and can't change your settings, budgets or guardrails. The only writes are a pending key request and a one-time setup code, and nothing is created without your approval. Sign-in uses OAuth with your AtlasBurn account, and access is limited to your own workspace's data.