If you're running AI agents in production, you already know the problem: costs are invisible until they're not. A single runaway agent loop can burn through hundreds of dollars in minutes, and by the time you notice, your OpenAI bill looks like a phone number.
AgentBurn is an open-source cost intelligence platform built specifically for AI agent infrastructure. It tracks every dollar across your entire stack — LLM API calls, compute sandboxes like E2B, tool integrations like Composio and Browserbase, and anything else your agents touch.
Why Agent Costs Are Different
Traditional cloud cost tools like AWS Cost Explorer or Vantage were built for infrastructure — VMs, storage, network egress. They don't understand the unit economics of AI agents:
- Token-level granularity — An agent might make 50 LLM calls per task, each with different input/output token counts and pricing tiers
- Multi-provider stacks — A single agent workflow might hit OpenAI for reasoning, Anthropic for analysis, and E2B for code execution
- Non-linear scaling — Agent costs don't scale linearly with users. A complex query might cost 100x more than a simple one
- Cascading failures — Retry loops, hallucination-triggered re-runs, and tool call chains can multiply costs exponentially
How AgentBurn Works
AgentBurn sits between your agents and your wallet. You instrument your agent code with a lightweight SDK or REST API call, and AgentBurn records every cost event in real time.
curl -X POST https://agentburn.dev/api/ingest \
-H "x-api-key: YOUR_KEY" \
-d '{
"agentId": "research-bot",
"provider": "anthropic",
"model": "claude-sonnet-4-20250514",
"operation": "llm_call",
"inputTokens": 2400,
"outputTokens": 800,
"costUsd": 0.08
}'From there, you get real-time dashboards showing cost breakdowns by agent, provider, model, and time period. Set budget alerts so you know before you blow past your monthly spend target.
Open-Core Model
AgentBurn's core is MIT-licensed. Self-host it forever, modify it, contribute back. The hosted Pro and Enterprise tiers add team features, higher event limits, and integrations — but the cost tracking engine itself is free and open.
Getting Started
Clone the repo, run the setup script, and you'll have a cost dashboard running locally in under two minutes. Point your agents at the ingest API and start seeing where your money goes.