juntoku9/sessory

Sessory is a tool for developers who use AI coding assistants and want to keep a record of what happened during each session. When you finish a session with Claude Code, a hook fires automatically, reads the session transcript, and saves a structured summary to a local database on your machine. The idea is that git commits show what changed in the code, but they do not show the AI's reasoning, the errors it hit, or how it corrected course. Sessory captures that information and makes it searchable. After a one-time setup using a single npm command, the capture runs on its own every time a session ends. The process has four steps: it reads the transcript, stores a row in a local SQLite file at a fixed path on your computer, calls a background AI classification step that extracts a structured summary, and then makes all of that available through a local web dashboard. Nothing is sent anywhere, the data stays on your machine by default. Each saved session includes a label for what kind of work it was (bug fix, refactor, test writing, and so on), a status indicating whether the work was completed or abandoned, the files touched, commands run, errors encountered, and a structured insight field covering what the bottleneck was and how problems were corrected. A significance score between 0 and 1 lets you filter out trivial sessions. The dashboard has four tabs: a session browser with full traces, a skills tab showing which AI slash commands were used across sessions, a table view for sorting and filtering everything, and a settings panel. You can search by file, command, error text, or tags. For teams, Sessory supports swapping the SQLite backend for a shared Postgres database by setting a single environment variable. That lets multiple developers write their sessions to one place so the shared archive is searchable across the team. Per-developer access control and secret redaction are listed as future work.