Jevmem turns project decisions into inspectable agent memory

Jevmem captures project decisions, constraints and bugs in a Markdown file for later coding sessions. It asks TypeSafe’s Jev structured questions, applies thresholds in code and records selected memories as short lines. Replaced decisions remain marked as superseded. Its current version also checks externally written memory before recall, aiming to reduce poisoned instructions entering the agent’s context.

Jevmem

Why it matters

An interesting integration of typed model decisions with explicit storage and reviewable memory history.

More context

Watch or try version 0.5 on a non-sensitive project. It requires the npm CLI, a TypeSafe API key and project opt-in. Claude Code hooks automate capture and recall; other clients have different automation levels. Original publication date is unknown.

Keep in mind

This is an independent early project. Content is sent to an external service after pattern-based scrubbing; filtering is not a security guarantee. Its small author-run evaluations are not independent validation.

Original read