See how hard you
ship with AI.
The public leaderboard for AI coding token usage. Every token your agents burn — Claude Code, Codex, Cursor, your own — counted, ranked, and put on a board against everyone else plugged in.
Paste this into your agent. It handles the rest.
Your agent fetches the setup instructions, reads the token counts already sitting on your disk, shows you the number, and asks before anything is published.
Set up burnlog for me — the public leaderboard for AI coding token usage. Fetch https://burnlog.net/agent-setup.md and follow it top to bottom. It is written to be executed, not summarised. Rules: show me my own totals before any account exists, ask me before anything is published, and never upload prompts, code, file names, or repo names — token counts only. I use Claude Code, which writes its own usage log, so the plain sync path covers it.
Sixty seconds, then never think about it again.
Finds your agents, signs you in, syncs. No config.
Counts calls no log file ever sees.
Every sync moves you up the public board.
If it burns tokens, burnlog counts it.
These write their own usage to disk, so burnlog reads it straight off — no key, no proxy, and your first run counts the history you already have.
No usage log to read, so these run under burnlog wrap, which counts calls no log file ever sees across 14 providers. No certificates, no HTTPS interception. What each agent supports →
The CLI is MIT-licensed and open source — read exactly what it sends before you run it. Or read the privacy model.
What people ask before they install it.
Does burnlog see my prompts or my code?
No. burnlog stores token counts, the model name, which agent produced them, an opaque dedupe id, and a timestamp. Prompts, completions, file names, file contents, repo names, working directories, and session ids are not part of the schema, so there is nothing to leak. The CLI is MIT-licensed, so you can read exactly what it sends before you run it.
Which AI coding agents does it track?
Claude Code, Codex, and opencode are read straight from their local session logs. Everything else is covered by burnlog wrap, which counts any command that talks to Anthropic, OpenAI, Google, Mistral, Cohere, OpenRouter, Groq, xAI, DeepSeek, Together, Fireworks, Perplexity, Cerebras, or Ollama over their standard base URLs — including agents that keep no readable log at all.
How does burnlog count tokens?
Two ways. It reads the usage numbers your agent already wrote to disk, and it reads the usage field off API responses for anything you run under burnlog wrap. The ranked number is input + output + cache writes; cache reads are stored and priced but not ranked, because every agent turn re-reads the whole cached prompt and counting that made the board a measure of session length rather than work. Cached tokens are handled honestly: OpenAI reports cached reads inside prompt_tokens, so they are subtracted before bucketing, and Anthropic splits usage across two streaming events, so both are collected before an event is counted.
Is burnlog free?
Yes. The leaderboard, ranks, achievements, and the README badge are free, the CLI is open source under the MIT license, and the whole app is self-hostable against your own Postgres. No card, no trial.
How do I get on the leaderboard?
Run npx @sxnalabs/burnlog. It finds your agents, shows you your numbers before asking for an account, signs you in with GitHub, syncs, and installs a hook so it keeps counting. Sixty seconds, no config file.
What are the ranks?
Eleven, on a logarithmic ladder: Spark, Ember, Blaze, Inferno, Supernova, Quasar, Singularity, Event Horizon, Heat Death, Vacuum Decay, and Boltzmann. Spark starts at zero and Boltzmann starts at ten trillion tokens, which nobody has reached. Your rank is derived from total tokens burned, so it updates itself on every sync and travels with you onto your profile page and your README badge.
Can I put my rank on my GitHub README?
Yes. Every account gets an SVG badge that updates itself, plus a dependency-free widget under 4KB for a portfolio or docs site. Profile links also unfurl into generated preview cards on X, Slack, Discord, and iMessage.
Does burnlog work with a team?
The public board ranks individuals today. Shared boards, budgets, and per-service API keys are built and running on a staging deployment while their UX is settled, so they are coming back to burnlog.net rather than being planned.