Open source · MIT licensed

Your agents forget This fixes that.

Hippocampus synthesizes a rolling briefing from your agent's daily notes and injects it into every session automatically. One command to install. Zero config.

$
View on GitHub Read the docs
Scroll
The problem

Every session starts
from zero

Your agent finishes a productive session. Threads identified, commitments made, context established. Then the session ends.

Tomorrow, it wakes up blank. No memory of what you discussed, what's pending, or what it promised. You re-explain. It re-asks. Commitments quietly disappear.

The more agents you run, the worse it gets. Multiply that amnesia across a team of 5, 10, or 15 agents and you're spending more time re-briefing than working.

The fix

A briefing,
not a brain dump

Hippocampus reads your agent's recent daily notes and synthesizes a rolling briefing: what's top of mind, what threads are open, what commitments are outstanding, what happened in recent sessions.

This briefing is automatically injected into every session. No tool call needed. Your agent just knows.

How it works

Three layers, one install

Layer 01

Daily Notes

Your agents already write memory/YYYY-MM-DD.md files. Hippocampus reads these. No changes needed.

Layer 02

Rolling Synthesis

A daily cron job reads the last 14 days and synthesizes HIPPOCAMPUS.md, a concise ops briefing loaded into every session.

Layer 03

Permanent Memory

When a pattern persists across synthesis cycles, promote it to MEMORY.md. The important stuff graduates. The noise doesn't.

The difference

Same agent, different experience

Without Hippocampus
You
"What were we working on yesterday?"
Agent
"I don't have context from previous sessions. Could you remind me?"
You
"We were debugging the auth middleware..."
With Hippocampus
You
"How's the auth middleware going?"
Agent
"Based on yesterday's session, the token refresh bug is narrowed to the middleware chain. You asked me to check the session store next. Want me to continue?"
Scale

One agent or thirteen.
Same install.

Running a single agent? Install and forget. Zero config required. Running a fleet? Configure each agent's synthesis independently: domain framing, custom sources, tailored output sections.

hippocampus.config.json 
{
  "agents": {
    "cyclawps": {
      "domainFraming": "Platform-centric. Lead with system health.",
      "sources": [{ "id": "health", "path": "artifacts/reports/health-check/" }]
    },
    "bobo": {
      "domainFraming": "Strategy-focused. Lead with pipeline and decisions.",
      "sources": [{ "id": "charter", "path": "CHARTER.md" }]
    }
  }
}
Output

Your agent's daily briefing

Section What it captures
Top of Mind 3-5 highest-priority items, rewritten every sync
Open Threads Active work streams, persists until resolved
Commitments Who owes what to whom, with staleness flags
Recent Sessions Rolling conversation log from the last 14 days

Custom sections available per agent. Add cron_health for your platform engineer, pipeline_status for your sales agent, whatever fits the role.

Compatibility

Works with what you have

Using lossless-claw?

Hippocampus is not a context engine. It uses before_prompt_build hooks; lossless-claw respects these during context assembly. No conflicts.

Using memory-core?

Hippocampus writes standard markdown files. memory-core indexes them automatically. They complement each other.

Using something else?

Hippocampus only reads and writes markdown in agent workspaces. It doesn't touch your config, your sessions, or your other plugins.

Provenance

Battle-tested,
not theoretical

Built by some dude that talks to 13 agents across a 4 business unit OpenClaw instance in production.

The three-layer memory pattern emerged from running autonomous agents across GTM advisory, e-commerce, content, and platform engineering, where dropped context means dropped revenue.

0
agents running
0
business units
0
tests passing
0
days rolling
FAQ

Questions

Hippocampus is a free, open-source OpenClaw plugin that gives your AI agents working memory between sessions. It reads daily notes, synthesizes a rolling 14-day briefing, and injects it into every new session automatically. Your agents wake up knowing what happened yesterday instead of starting blank.
MEMORY.md is permanent storage: facts that never change. But what about last week's open threads? Yesterday's commitments? The thing someone asked you to follow up on? That's what the rolling synthesis handles. HIPPOCAMPUS.md is a 14-day briefing that stays current. MEMORY.md is forever. You need both.
Yes. Install, add a cron job, done. No multi-agent config required. The defaults work.
Default target: 3,000–5,000 characters. Roughly one page. Big enough to be useful, small enough to leave room in the context window.
Yes. Add custom sources (any file or directory in your workspace), custom output sections, and domain framing per agent. Or use the defaults. They work for most setups.
No. It's an LLM skill. The model reads your sources and writes a briefing following structured instructions. The output varies slightly between runs, but the sections and content sources are consistent.
The HIPPOCAMPUS.md is injected via prependSystemContext, which benefits from provider prompt caching (stays static between turns). At 3–5k characters, it's a small fraction of a typical context window.

One command.
That's it.

$ openclaw plugins install hippocampus

Free. Open source. MIT licensed. No account, no API key, no sign-up.

Or ask your agent: "Set up hippocampus for me"

View on GitHub Read the docs