Memories

What memories are

Memories are short, durable notes scoped to a project. Unlike a single chat — which the model forgets when you close it — memories are stored with the repo and injected into agent and chat context every time, so the assistant keeps remembering what matters: a gotcha you hit, a convention you settled on, the command that actually builds the app.

How they’re stored

Memories live in a plain Markdown file at .goatfied/memories.md in your workspace root, one note per bullet. Because it’s just a file in the repo, it’s human-editable, reviewable in a pull request, and shared with everyone who clones the project.

# Goatfied memories

- The dev server runs on port 5180, not 3000.
- Never edit files under src/generated/ — they're built from protos.
- Prefer pnpm; this repo's lockfile is pnpm-only.

Adding and removing

In the assistant panel, type a slash-command to manage memories without sending anything to the model:

• /remember <note> — append a memory (duplicates are ignored).
• /forget <note> — remove a memory by its exact text.

You can also edit .goatfied/memories.md directly — the assistant picks up changes on the next message.

How they’re used

On every request, Goatfied reads .goatfied/memories.md and prepends it to context alongside your rules. Rules encode standing conventions; memories capture project-specific facts you discover along the way. Keep them short — a focused list steers better than a sprawling one.

Memories vs. rules

• Rules are deliberate, authored guidance (“always use async/await”).
• Memories are quick, incremental facts you capture mid-flow (“the staging DB is read-only”).