Hippo + OpenClaw Integration

April 16, 2026 ยท View on GitHub

Install the native OpenClaw plugin directly from npm:

openclaw plugins install hippo-memory
openclaw plugins enable hippo-memory

This loads Hippo as a real OpenClaw plugin instead of relying only on prompt hooks. The plugin resolves memory from the active workspace by default, so it uses the workspace .hippo/ store and automatically merges your shared ~/.hippo/ memory during recall and context.

If you enable autoSleep in the plugin config, OpenClaw session-end consolidation now runs in a detached background worker. The session can end immediately while hippo sleep finishes in the background.

Daily cadence is separate. hippo init registers the workspace and installs a single machine-level daily runner that sweeps every registered Hippo project at 6:15am. OpenClaw retrieval stays query-driven: the current workspace store and the shared global store are searched together.

Recommended split:

  • autoContext: true for session-start retrieval
  • autoLearn: true for filtered in-session error capture
  • autoSleep: false if you prefer a daily scheduler and zero session-end work
  • autoSleep: true if you want per-session consolidation without blocking exit

If your project has a .openclaw directory or AGENTS.md, hippo init auto-detects and patches AGENTS.md with the memory hook. No extra steps.

cd my-project
hippo init
# Auto-installed openclaw hook in AGENTS.md

Manual install

hippo hook install openclaw

Use this hook path if you want lightweight prompt injection without installing the native plugin.

This appends a <!-- hippo:start --> ... <!-- hippo:end --> block to AGENTS.md:

## Project Memory (Hippo)

At the start of every session, run:
hippo context --auto --budget 1500

On errors or unexpected behaviour:
hippo remember "<description of what went wrong>" --error

On task completion:
hippo outcome --good

After significant coding sessions:
hippo learn --git

To remove: hippo hook uninstall openclaw

OpenClaw Skill (alternative)

If you prefer using OpenClaw's skill system instead of patching AGENTS.md, create this file:

~/.openclaw/skills/hippo/SKILL.md

# Hippo Memory Skill

Biologically-inspired memory for AI agents. Decay by default, retrieval strengthening, sleep consolidation.

## When to activate

Always. Every session.

## Session start (MANDATORY)

Before doing any work, load relevant context:

hippo recall "<task description from user message>" --budget 3000

Use the output as direct context. Do not summarize or skip it.

## During the session

When you learn something worth keeping:
hippo remember "<lesson in 1-2 sentences>"

When you hit an error or discover a gotcha:
hippo remember "<what failed, why, how fixed>" --error

When the user says "remember this" or "save this":
hippo remember "<what they said>" --pin

## Session end

After completing work:
hippo outcome --good   # if recalled memories helped
hippo outcome --bad    # if recalled memories were irrelevant

If 10+ new memories were created this session:
hippo sleep

Multi-repo git learning (cron)

Set up a weekly cron to scan all your repos:

hippo learn --git --repos "~/repo1,~/repo2,~/repo3" --days 7
hippo sleep

This extracts lessons from fix/revert/bug commits across all repos and consolidates them.

Global memory

To share memories across all projects:

hippo init --global
hippo remember "API key rotation happens quarterly" --global
hippo recall "api keys" # searches local + global automatically