OpenViSTAI · Agents · Blockchain · Crypto
Menu
Agent Memory & RAG

Hipocampus

Hipocampus is a JavaScript, MIT-licensed memory harness for AI agents that stores persistent context in files and organizes it into hot, warm and cold tiers. Its design combines an always-loaded topic index, working-state notes, on-demand records, hybrid search and a hierarchical compaction tree. The README names Claude Code, OpenCode and OpenClaw as compatible environments.

Independent OpenViST project profile, based on linked sources.

Hipocampus — OpenViST project profile in Agent Memory & RAG.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Hipocampus when an agent workflow needs to carry context across sessions and connect a current task with earlier discussions that may not share its keywords.

Who it is for

  • Developers configuring persistent memory for Claude Code, OpenCode or OpenClaw agents.
  • Agent builders who want past decisions and topics surfaced even when a new task does not include obvious search terms.

Documented uses

  • Keep an overview of topics from past agent sessions available in a compact index.
  • Retrieve specific details from daily logs, curated knowledge and plans when needed.
  • Organize long-term context through hierarchical compaction and search.
Before you choose

Requirements & limits

The repository is public and declares an MIT license. The README documents npm initialization and a Claude Code plugin setup, and names Claude Code, OpenCode and OpenClaw as supported environments.

  • ROOT.md is described as an always-loaded topic index, with a configurable token budget; the documented default is 3,000 tokens.
  • Vector embeddings are enabled by default and listed as using about 2 GB of disk; the README documents a BM25-only option.
  • The documented memory layers distinguish always-loaded work context, on-demand records and search or compaction-based retrieval.
  • The project reports benchmark results on MemAware, but those figures describe that evaluation and do not guarantee results for a different agent or task.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Initialize Hipocampus from npm

Set up the standalone Hipocampus harness using the documented npm initialization command.

Before you start

  • An environment with a supported agent platform: Claude Code, OpenCode or OpenClaw.
  • The README documents initialization through npx.

Steps

  1. Run the command below.

    Documented command
    npx hipocampus init
  2. If using Claude Code, OpenCode or OpenClaw, the README says platform detection is automatic; its configuration also documents a platform override.

Expected outcome

The documented initialization route sets up the standalone memory harness.

Guide limitations

  • Vector embeddings are enabled by default and are listed as using about 2 GB of disk; the documented --no-vector option selects BM25-only search.
Source-based practical guide

Set up the Claude Code plugin

Add the Hipocampus plugin through Claude Code and run its documented initialization step.

Before you start

  • Claude Code.
  • The README's plugin marketplace and installation commands.

Steps

  1. Run /plugin marketplace add kevin-hs-sohn/hipocampus.

  2. Run /plugin install hipocampus@kevin-hs-sohn/hipocampus.

  3. Run the command below for full setup.

    Documented command
    npx hipocampus init

Expected outcome

The documented plugin installation is followed by the full setup initialization step.

Guide limitations

  • The README does not include further instructions for using the configured memory workflow.
Evidence and transparency

Sources & review

Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
kevin-hs-sohn/hipocampus · Official homepage title · retained in its source language