Hippo Memory

Hippo is a long term memory layer for AI coding agents. It stores lessons from your sessions locally and surfaces them again in later ones, so an agent does not repeat the same deployment mistake or forget a project convention such as which package manager a repository uses. The design borrows from how human memory behaves. New information lands in a session buffer, becomes an episodic memory that decays on a one year half life unless it is used, and gets stronger each time it is recalled. Failures can be tagged so they decay more slowly and resurface when you return to the relevant code, and you can mark a recalled lesson as wrong so it stops coming back. A sleep step merges three or more related episodes into a single semantic pattern. Commands include capture of errors, outcome marking and a doctor check that verifies the install. The vendor is candid that decay and sleep have not been shown to improve recall in its own tests, and that outcome marks and retrieval strengthening are what measured as helpful. Hippo connects through hooks and MCP to tools such as Claude Code, Codex, Cursor, OpenCode and other MCP clients, and keeps proprietary data on your machine. It is aimed at developers who run coding agents across multiple tools and sessions. The vendor reports 98.0 percent recall at five on the LongMemEval S benchmark with a free local embedder. Installation is through a global npm package, and the site lists no paid plan or price.