atlasburn

See what each AI feature of your Python app spends, and stop runaway agent loops before they hit your bill.

atlasburn points your existing OpenAI, Anthropic or other OpenAI-compatible client at the AtlasBurn Edge Proxy in one line. Every call is then tracked by model and feature, and protected by the guardrails you set in AtlasBurn: alerts, throttling, and a kill switch that pauses only the feature that is looping.

pip install atlasburn

Use it

Get a key at app.atlasburn.com (Settings → API keys) and set ATLASBURN_KEY.

import atlasburn
from openai import OpenAI

client = OpenAI(**atlasburn.openai_config(feature="support-chat"))

client.chat.completions.create(model="gpt-4o-mini", messages=[{"role": "user", "content": "Hi"}])

Anthropic:

from anthropic import Anthropic

client = Anthropic(**atlasburn.anthropic_config(feature="summarize"))

Tag a single call with a different feature:

client.chat.completions.create(..., extra_headers=atlasburn.feature("search"))

Other OpenAI-compatible providers: atlasburn.openai_config("openrouter"), "mistral", "groq", "together", "deepseek", "xai". Your provider key is used exactly as before.

Safe by default

• No dependencies. It only builds the client's base_url and headers.
• Never breaks your app. Without a key it warns and returns an empty config, so your client talks to the provider directly (untracked).
• Content-blind. AtlasBurn records usage (model, tokens, cost, timing, your feature label), not prompts or outputs.

What's next

This first release covers the proxy. A full Python SDK with automatic capture (no base-URL change, like the Node.js SDK) is in progress under this same package name.

Docs: Proxy quickstart · Guardrails and the kill switch · atlasburn.com

Apache-2.0 © 2026 AtlasBurn