Codicent

Codicent Muse gives every coding repo one shared, file-first memory. The problem: every AI coding session starts from zero. Decisions evaporate, tacit knowledge ("why we did it this way") lives in someone's head, and project history gets stuck in one vendor's private schema. The next agent, even a smarter one, starts over. What we built: a small kit that adds a Codicent logbook to any repo, as plain Markdown + YAML in git under .codicent/. It contains: - Idempotent installers (install.sh / install.ps1) that never overwrite existing entries - A memory contract (spec): what to log, when, and how (decisions, releases, tacit knowledge) - A starter logbook with index and templates (#decision, #release, #tacit) - A 10-line AGENTS.md snippet that teaches the coding agent to read the logbook at session start and write an entry after it acts - A docs site Muse Code uses it with zero extra tooling. Read: in a fresh session, "Read the repo logbook and summarize the latest decisions" returns real entries with dates and links. Write: given an ordinary request in a demo app ("switch the config file from JSON to TOML"), Muse Code made the change and, without being asked, wrote a #decision entry (context, three options, trade-off, consequences) plus an index line in the same commit. Both runs are shown in the demo video. The repo dogfoods itself; its own .codicent/ logbook records every decision and release made while building it. What makes it different: the memory is owned by git, not by a vendor. It survives session changes and vendor switches, and any future model gets full use of the repo's history. Next step: the same entries flow into Codicent, the common memory for repos, ideas and projects, which calls Meta's models (Muse Spark 1.3) through the Meta Model API. Open source (MIT). Links: - Repo: https://github.com/izaxon/codicent-muse - Docs site: https://izaxon.github.io/codicent-muse/ - Demo video: see the video field