Skip to main content

Loom

The history of a code project, and the conversations that made it.

Loom reads a git repository and draws it on one timeline. Each area of the code gets a line. Each commit is a point. When an agent conversation made a commit, Loom ties the two together.

The cartridge id is @mnemosyne-plugins/loom. Its code lives at Mnemosyne-OS/Loom, under the MIT licence. Loom runs inside Mnemosyne OS only.

First steps​

  1. 1. The repository

    Press Choose a repository and pick a git repository folder. Loom reads dates, authors, commit messages and file names. It never reads the content of your files.

  2. 2. The agent conversations

    Press Choose the folder and add the Claude Code projects folder (usually .claude/projects in your home folder), or any folder of conversations. Loom ticks the folders that look like this repository. Tick others by hand. Without conversations, Loom draws the commits only.

  3. Read

    Press Read. Loom shows its progress, and Stop ends the reading. A stopped reading draws nothing from the partial pass.

The next reading is faster. Loom keeps a summary of each unchanged conversation file, and the commits it already read.

Reading the timeline​

Drag to move, use the wheel to zoom, click a name or a point. Whole history fits everything on screen.

  • Lines switches between By app (one line per area of the code) and By category (one line per commit scope). Every scope shows all scopes. Busiest folds the small ones into "others".
  • A commit's card lists its files and its conversations. Made in this conversation means the conversation printed the commit. Probably made in means it wrote one of these files at that time: a coincidence, not a record.
  • A line's card shows Commits per week, Conversations that worked on it and Files changed most often. Split into sub-lines opens an app into its folders.
  • Milestones marks the numbered design docs and the agent memory notes on the timeline.

The header counts the commits and conversations, the share of commits recorded, probable or without a conversation, and the time of active conversation.

Summary​

A line's card offers Prepare a summary. Loom first says how many characters it will send to your active model. It sends the commit messages and what you typed in the tied conversations, never the agents' replies. Press Send to write it. The cost appears in the cost journal of Mnemosyne OS.

Where your data goes​

  • Loom only reads the repository and the conversation folders.
  • The reading cache lives in the cartridge's own browser storage. If it is lost, Loom reads everything again.
  • Open on a conversation writes it as a readable document beside its transcript, on your press.
  • Loom holds up to 16 repositories. Forget this repository removes one.

For agents​

Loom exposes 16 actions.

  • Read: status, lines, commits.
  • Write: read, openCommit, openLine, setLineMode, fit, showMilestones, openMilestone, closeView, stopReading, showSources, expandLines, chooseRepository, tickProject.

read is long: it answers started, then follow with status. Before the first reading the board answers NOT_READ. chooseRepository only switches to a repository the person already gave Loom.

Voice examples: "show the latest commit in Loom", "show the milestones", "close the card".