Your prompts don't leave your machine.

The burnlog CLI reads local log files written by your coding agents — e.g. ~/.claude/projects/*/*.jsonl for Claude Code, ~/.codex/sessions/**/*.jsonl for OpenAI Codex. These files contain token usage alongside prompts, tool calls, and file contents.

The CLI extracts only the token-usage numbers and a random request id (for dedup). Everything else — prompt text, filenames, working directory, session ids, tool output — is discarded locally and never transmitted.

Per burn event, we store exactly these fields:

• requestId — a random opaque id used only to detect duplicate uploads.
• source — which CLI (claude-code, codex, …).
• model, provider — e.g. claude-opus-4-7, anthropic.
• inputTokens, outputTokens, cacheCreationTokens, cacheReadTokens, totalTokens.
• timestamp — when the model call happened.

We do not store: prompts, responses, file paths, project or repo names, working directory, session ids, tool outputs, or any content produced during your agent session.

The source-of-truth is the Prisma schema at web/prisma/schema.prisma in the public repo. If you can read it, you can audit it.

Signing in uses GitHub OAuth. We receive your GitHub username, display name, avatar URL, email (if public), and bio. We use these to show your profile on the leaderboard. You can delete your account at any time from settings; this cascades and permanently removes every burn event we've recorded for you.

We set a session cookie to keep you logged in. We don't run third-party analytics, pixels, or ad trackers. If we add a self-hosted analytics tool in the future, we'll list it here before turning it on.

burnlog is open source (MIT). You can self-host the entire stack — web + Postgres — and keep the data on your own machine. Point the CLI at your instance with BURNLOG_API_URL.

burnlog is operated by Sharvil Saxena / SXNA Labs. For privacy questions, email [email protected].