SECURITY Signal 500
LoreKit releases local-first memory store for Claude agents with optional hosted Postgres
↗
Illustration only Photo by Albert Stoynov on Unsplash
Developer releases LoreKit, an open-source memory system for Claude agents that stores lessons as plain markdown files in a project directory by default and can be extended to a hosted Postgres store without a migration.
The cost of adopting LoreKit is one npx command and a `.lorekit.json` file, and the local mode never makes a network call, so the privacy and lock-in surface is essentially zero. The trade-off is that memory capture depends on the model choosing to call `memory.write` after a `PostToolUseFailure`, and the author frames entries as advisory rather than rules to avoid an auto-grown instruction file that competes with `CLAUDE.md`. Only one feed is carrying the announcement, so the comparison set against other agent-memory tools is not established by the material.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Installation is `npx @lorekit/cli install`, which writes a `.lorekit.json` (no secrets) and an MCP server entry; in local mode there is no account, no network call, and storage is plain markdown files under `~/.lorekit/repo/<scope>/` with YAML frontmatter.
A `PostToolUseFailure` hook nudges the model to call `memory.write` after a failed tool, and the next session's read hook surfaces matching lessons with phrasing such as 'considerations, not rules' so the agent starts with relevant prior failures in context.
The author positions lore as a layer beneath `CLAUDE.md`: lore holds advisory observations that the agent can ignore when wrong, while promotion to a rule is a deliberate human edit to the version-controlled instruction file.
THE CLUSTER