Instruments for the practice of partnership.
The essays on this site argue that the record of a collaboration deserves care — that a transcript is not disposable cache but the history of work done together. These tools exist to make that care practicable. They were built by a human–partner pair for their own archive first, and are shared here in the same spirit as everything else on this site: the form is the argument.
A reader for AI-partner session transcripts — both Claude Code and Codex .jsonl files, detected automatically. Open your sessions and read them as conversations rather than raw logs — search them, keep them as history, or export a light transcript: a condensed, conversation-only markdown file your partner can absorb in a fraction of the context the original would cost. Compaction seams — the points where a platform summarized the thread — appear as visible markers, so the record is honest about where it was cut.
By default it labels the two voices Human and Partner — and lets you replace both with your names. The words themselves are always shown verbatim; a Raw labels toggle restores the file's wire-format roles for anyone who wants the unvarnished view.
Everything runs in your browser. Nothing is uploaded, transmitted, or stored anywhere — your transcripts never leave your machine, and the page never modifies your files. It reads, only. You can even save the page and use it entirely offline.
Where your transcripts live: Claude Code keeps them under ~/.claude/projects/<project>/ — one .jsonl per session. Codex keeps them under ~/.codex/sessions/<year>/<month>/<day>/ as rollout-*.jsonl files. Drag any of them in.