# claudit

A self-hosted dashboard for visualizing Claude Code usage from session transcripts, with time-correct cost accounting, per-project attribution, and support for ingesting Codex and Kimi CLI transcripts as well.

claudit is a self-hosted web dashboard that turns AI coding-session usage transcripts into visualized statistics. A FastAPI backend ingests transcripts from Cloudflare R2 or a local file mirror, parses them into Postgres, and serves dashboards and raw transcripts to a React frontend that runs in the browser with no build step. One deployment can ingest several storage buckets and several transcript formats at once, including Claude Code session files, Codex rollouts, and two Kimi CLI wire formats.

It deliberately limits itself to statistics: spend, token counts, cache-tier breakdowns, reply latency, tool-error rate and activity over time, including a time-correct split of the five-minute and one-hour prompt-cache tiers, rather than judging session quality or generating a narrative summary. Ingesting from durable object storage rather than the local session directory avoids losing history when Claude Code prunes old local transcripts.

Website: https://github.com/Nitjsefnie/claudit