Hippo Memory

Make your agent's memory work like a brain. Hippo is long-term memory for coding agents. Memories decay over time unless retrieved. Errors stick longer. Sleep consolidates.

At the start of a task

The plugin's UserPromptSubmit hook injects pinned rules and the memories that match each prompt, so do not load them yourself.

Only if no hippo memory block has appeared in this session, run this at the start of a task:

hippo context --auto --budget 1500

Read the output. These are things the project has learned from past sessions.

When you learn something non-obvious

hippo remember "<the lesson in 1-2 sentences>" --tag <category>

When you hit an error or unexpected failure

hippo remember "<what failed, why, and how you fixed it>" --error

The --error flag doubles the half-life. Hard lessons don't fade quietly.

After completing work successfully

hippo outcome --good

If the recalled memories were irrelevant:

hippo outcome --bad

When stuck or hitting repeated errors

hippo recall "<what's going wrong>" --budget 2000 --why

When making an architectural decision

hippo decide "<decision>" --context "<why>"

Decisions get a recall boost so they surface when relevant, and stay active until superseded.

When migrating or replacing a tool

hippo invalidate "<old tool>" --reason "migrated to <new tool>"

Or let hippo learn --git detect it automatically from migration commits.

After significant coding sessions

hippo learn --git

This scans recent commits, auto-extracts lessons, and invalidates memories about replaced tools.

Export memories

hippo export memories.json
hippo export --format markdown memories.md

Check memory health

hippo status