cc-ledger captures every Claude Code, Cursor, and Copilot session in your repos and turns it into per-user, per-repo metrics — token cost, line attribution, throughput, streaks. So engineering leaders know what their agents are actually doing.
Per-model token usage in dollars. Cache reads at 10% rate, batch tier auto-applied. Roll up by week, by team, by repo.
Reads commit trailers and diffs to figure out which lines were authored by an agent and which by a human. Per file, per PR, per repo.
Where work converts. 14k prompts → 2k commits attributed → 312 PRs touching AI code → 81% acceptance. Surface the leaks.
Which engineers haven't started. Which repos are AI-heavy. Daily streaks, session length, message volume — per user, per repo.
Directors, VPs, and team leads need to answer one question: is the AI spend justified, and which teams get leverage?
cc-ledger is the dashboard, not the raw logs. No spelunking through JSONL transcripts. Open the page, see cost, attribution, throughput. Drill into any user, any repo, any week.
One command wires Claude Code lifecycle hooks into your repo. No daemons, no SaaS pings — everything runs locally.
$ cargo install cc-ledger && cc-ledger installEvery prompt, every tool call, every diff is recorded into a local SQLite ledger and (optionally) committed to a private git branch alongside your code.
Authorize the GitHub App, point cc-ledger at your repos, and the dashboard fills in real time. Cost, attribution, throughput — per user, per repo, per week.