The patterns between sessions

AI coding sessions produce a great deal of evidence (decisions, reversals, repeated mistakes, useful interventions), but most of it disappears into transcripts after the work is done.

Tessera reads local Claude Code, Codex, and Gemini CLI traces and turns them into a weekly retrospective. It builds a narrative for each session, looks across them for recurring patterns, and presents the findings in an interactive HTML dashboard where each observation can be rated.

Closing the loop

The second half of the project is a lightweight live coach. It watches for a small set of mechanical waste patterns during a Claude Code session and uses the retrospective ratings to decide which nudges are worth surfacing.

The idea is not to add another agent that tells someone how to work. It is to make reflection cumulative: evidence from completed sessions shapes what gets noticed in the next one, while the underlying traces remain on the user's machine.