Logpile
local-first · Claude Code + Codex

Agents do the work. Logpile keeps the record.

Every session your coding agents run is work product — and it's sitting in JSONL files you'll never open. Logpile indexes all of it into a searchable local archive: transcripts you can read, filter by what they did, and publish with secret scanning.

$ git clone https://github.com/MaxGhenis/logpile && cd logpile $ ./logpile.sh sync && ./logpile.sh serve

Local-first. Nothing leaves your machine unless you publish it. MIT-licensed.

27,918 sessions · 632M output tokens · 974 repos — the author's index, as of Jul 16, 2026.

The record — 14 dayslive index · Jul 16, 2026
2026-07-16 8M out 758 sessions
2026-07-15 7M out 259 sessions
2026-07-14 8M out 328 sessions
2026-07-13 12M out 231 sessions
2026-07-12 18M out 537 sessions
2026-07-11 59M out 803 sessions
2026-07-10 53M out 765 sessions
2026-07-09 12M out 404 sessions
2026-07-08 15M out 395 sessions
2026-07-07 15M out 238 sessions
2026-07-06 13M out 333 sessions
2026-07-05 17M out 444 sessions
2026-07-04 6M out 175 sessions
2026-07-03 5M out 81 sessions

01 · keepEvery session, indexed

One command turns the session files Claude Code and Codex leave on disk into a SQLite archive — with token accounting that dedups Codex fork replays and cross-file resumes, so the numbers are right.

02 · readTranscripts, not log soup

Sessions render as readable transcripts you can search and filter by what they actually did — wrote files, ran tests, committed — and by origin: your work vs. delegated agents vs. pipelines.

03 · publishProof, when you choose

A review queue scans every transcript for secrets and PII before it can go public. Published sessions build an operator profile: the receipts for how you actually work with agents.