Jevmem – automatic project memory for Claude Code, built on Jev

Your coding assistant's memory becomes persistent and significantly faster for decision recall.

Jevmem, a new open-source tool built on Jev, automatically captures and stores decisions, constraints, bugs, and to-dos from coding assistant chats, such as Claude Code, Cursor, and Codex. It saves these to a JEVMEM.md file within a project, marking superseded information instead of deleting it. This memory is then added to the coding assistant's context for subsequent sessions. Installation involves adding a Claude Code plugin or using `jevmem init` for direct integration. Jevmem supports various AI models for decision-making, with its own `jevmem auto` decider benchmarked at 0.30 seconds per decision, significantly faster and cheaper than leading LLMs like GPT-6 Astra and Claude Opus 5.5. Privacy is maintained as only user messages and memory lines are sent to TypeSafe AI for processing, with sensitive information scrubbed.

Jevmem introduces an efficient and cost-effective method for maintaining persistent memory in AI coding assistants, directly impacting the development workflow by ensuring context is continuously retained and accessible. Its benchmarked speed of 0.30 seconds per decision, compared to several seconds for major LLMs, highlights a leap in real-time context management, crucial for complex coding tasks where quick recall of project specifics is vital.

This development signals a shift towards more integrated and intelligent agentic workflows, affecting developers who rely on AI pair programmers by providing a more coherent and less forgetful AI partner. The tool's ability to operate discreetly and affordably, even without an LLM key for its core decision process, makes advanced memory capabilities accessible across a wider range of projects and user preferences.