Show HN: Hippo, biologically inspired memory for AI agents

What is Hippo

Hippo is an open-source, biologically inspired memory layer for AI agents — think hippocampus for your bots. It stitches together memories across tools (Claude Code, Codex, Cursor, OpenClaw and any CLI agent), stores data in a SQLite backbone with markdown/YAML mirrors, and keeps everything git-trackable and human-readable. No runtime dependencies, just Node.js 22.5+; optional embeddings via @xenova/transformers for better recall. Tired of starting from zero every time you switch tools? You’re not alone.

How it works

Hippo isn’t a dump truck that hoards everything. It offers multi-layer memory (working vs long-term), decay mechanics, tags and confidence levels, hybrid BM25+embedding search, session handoffs, and explainable recall so you can see why something was brought back. Commands are refreshingly simple: install, hippo init, hippo remember "…" --tag error, hippo recall "…" --budget 2000. It auto-detects agent frameworks and can wire itself in; there’s even an auto-sleep hook for Claude Code to run cleanup when a session ends.

Why it matters

Developers who hop between agent platforms — or teams re-learning the same lessons — will feel the relief. Hippo aims to make useful memories stick and noise fade, not just pile up facts in a filing cabinet. It has been reported that hippo agents dropped from a 78% trap rate to 14% over a 50-task sequence, a dramatic claim that suggests the approach can change agent behavior if it holds up under wider testing. Want portability, visibility, and fewer repeated mistakes? Hippo is an intriguing, practical attempt to build memory that behaves a bit more like a brain than a storage closet.

Sources: github.com/kitfunso, Hacker News