Claude Transcripts — documentation

Self-hosted history for your Claude Code sessions. A hook logs every session to your own CouchDB + S3; a web API serves it back; a web UI, a CLI, and agents read it. Nothing leaves your network.

Work in progress — not tested as ready for use. No installation has been walked end to end on a clean machine. Breaking changes land without notice, stored data may need discarding between revisions, and there is no auth or security model (Tier 1 assumes one user on one trusted machine). These docs describe the intended design as much as the current state.

Start here

How these docs are organised

• Getting started — installing, configuring, and recording your first session. Written for someone running the project.
• Development — setup, conventions, tests, and the generators. Written for someone changing the code.
• Operations — releasing, containers, migrations, and application logging.
• Reference — per-component and per-surface detail: webapi, webui, CLI, the hook, routes, CouchDB documents.
• Design & specification — what the system is meant to be, the tier model, the roadmap, and the technology choices.
• Decisions — one ADR per architectural decision, nested under design.

Technical design is treated as a first-class deliverable here. This is a fresh rebuild: the design set was re-written rather than copied from the predecessor project, with naming, ports, paths, and reversed decisions reconciled to this repo's conventions.

The published site at vredchenko.github.io/claude-transcripts/docs renders the same tree with this structure as its navigation, and its index lists every page — generated from what actually shipped, so it can't drift.

Conventions

Naming, components, ports, and stack conventions are in conventions.md. Repo-level rules for agents working in this codebase are in CLAUDE.md.