hippo-memory
Local long-term memory for coding agents with an optional Jev reranker that orders recalled memories.
Overview
Hippo is a memory layer for coding agents such as Claude Code, Codex, Cursor, OpenClaw, OpenCode and Pi. It stores memories in a local SQLite database with Markdown mirrors, searches them with BM25 by default, and learns from outcomes: memories marked wrong drop in rank, frequently used ones get stronger, and superseded facts leave recall. hippo init installs hooks and AGENTS.md or CLAUDE.md instructions, and any MCP client can connect.
Jev is an opt-in reranker for recall, off by default. The README reports that on a private 300-query developer store R@1 rose from 0.41 to 0.62 with the Jev reranker compared with the free local cross-encoder, at about 0.0004 USD per recall. It also reports a negative result: on a 150-question LongMemEval set Jev did not improve answer rate, though two Jev-ranked memories answered as well as five cross-encoder-ranked ones, giving a shorter context.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How relevant is this stored memory to the current query?(inferred) | — |
How it works
- Agent issues a recall query through hooks, CLI or MCP
- BM25 (or embeddings) retrieves candidate memories
- With --reranker jev, Jev scores each candidate's relevance
- Top-ranked memories are returned into the agent's context
Strengths
- README reports R@1 0.41 to 0.62 on a private store with the Jev reranker
- Fewer memories needed in context for the same answer quality on LongMemEval
- Reranker is opt-in, so the default setup makes no network call
Caveats
- Jev reranker showed no answer-rate gain on LongMemEval per the README
- The headline improvement was measured on a private store
- Using the reranker sends memory text to the TypeSafe API
Getting started
`npm install -g hippo-memory && hippo init`, then `hippo recall "<query>" --reranker jev` to use Jev.At a glance
- Type
- Project
- Status
- production
- Language
- TypeScript
- Runtime
- Node CLI, hooks, MCP server
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Agent memory: a biologically-inspired memory store whose optional Jev reranker lifts recall R@1 from 0.41 to 0.62 on a private 300-query developer store.
yibie/awesome-jev
Frequently asked questions
What does Jev do in Hippo?
It reranks candidate memories for a recall query so the most relevant ones come first.
Is Jev required?
No. Search is BM25 by default; the Jev reranker is enabled per recall with --reranker jev.
What does it cost?
The README estimates about 0.0004 USD per recall when the Jev reranker is used.
Related projects
CLI and MCP server giving coding agents Jev-powered semantic search, focused reads and list picking in about a second for about a cent.
BenchmarkbetaPythonZilliz's cross-agent Markdown memory with an optional Jev reranker that asks yes/no relevance questions about retrieved chunks.
LibraryproductionPythonMulti-user MCP 'agents' cloud' on Cloudflare Workers with an optional search stage that reranks candidates with Workers AI Jev scores.
MCP serverbetaTypeScriptLocal-first MCP server that returns only Jev-checked passages from big PDFs, Word files and spreadsheets, with exact row calculations.
MCP serverbetaTypeScriptAgent skill and Python library for Claude Code and Codex that reranks candidate lists with one Jev yes/no relevance question each.
ProjectbetaPythonSelf-hosted memory MCP server on Cloudflare Workers where a Jev layer judges memory type, salience and durability and re-ranks recall.
MCP serverprototypeTypeScript