Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jevmem

Project memory for Claude Code that asks Jev whether each chat statement is worth keeping and checks tool calls against saved rules.

ProjectbetaMIT
View on GitHub ↗

Overview

jevmem saves the decisions, rules and failed approaches from Claude Code chats to a JEVMEM.md file in the repository and brings the relevant ones back in later sessions. Before Claude runs a command or edits a file, the call is checked against the saved rules. It is listed in Anthropic's Claude plugin directory and on the MCP Registry, and works with Cursor and Codex over MCP.

When you state something like 'we use Postgres', jevmem asks Jev a yes/no question about whether it is worth keeping; if so, an LLM writes it as one line. The README reports results from the author's own tests on three small projects: Claude followed earlier decisions in 66 of 72 sessions with jevmem versus 28 without, attempted forbidden changes in 0 of 18 versus 10 of 18, and a hand-maintained CLAUDE.md did about as well (67 of 72). Saving costs about 17 cents per 1,000 messages in the background; recall adds a third to half a second per prompt.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this statement a decision, rule or failed approach worth keeping?—
Yes / noDoes this command or edit conflict with a saved rule?(inferred)—
ChoiceWhich saved lines are relevant to this prompt?(inferred)—

How it works

  1. You chat normally in Claude Code
  2. Jev judges whether each new statement is worth remembering
  3. An LLM writes kept items as one line in JEVMEM.md
  4. Relevant lines are injected at the next session's prompts
  5. Tool calls are checked against saved rules before running

Strengths

  • Keeps a CLAUDE.md-style file up to date automatically
  • Saving runs in the background at low cost
  • Works with Claude Code natively and Cursor or Codex via MCP

Caveats

  • Results are the author's own test sets, not an independent benchmark
  • Recall latency and cost grow with the memory file
  • Needs a TypeSafe key and an LLM for writing lines

Getting started

Install from Anthropic's Claude plugin directory or npm (jevmem); see the docs site at avinash-jetwani.github.io/jevmem.

At a glance

Type
Project
Status
beta
Language
TypeScript
Runtime
Claude Code plugin, MCP server
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • A Claude Code memory manager that uses Jev to filter decisions and rules from chat history into durable, per-repo memory.

    logicrw/awesome-jev-projects

  • jevmem saves the decisions, rules and failed approaches from your Claude Code…

    wh000wh000/awesome-jev-live

  • Jev decides. The LLM writes one line. Your project never forgets. Jev-powered memory layer for AI coding tools.

    hellogumbo/awesome-jev

  • Automatic project memory for Claude Code, Cursor and Codex, with typed decisions choosing what is worth remembering.

    MrJev/awesome-jev

Frequently asked questions

What does Jev decide in jevmem?

Whether a statement from your chat is worth saving as project memory, as a yes/no probability.

Is it better than a hand-written CLAUDE.md?

The author's tests found a maintained CLAUDE.md did about as well; jevmem's value is keeping it updated for you.

Which tools support it?

Claude Code as a plugin, and Cursor and Codex over MCP.

Related projects

  • jev-awesome-skills

    Open-source Jev skills for Claude Code, Codex, Cursor and Grok: one client to gate, route, guard, verify, triage, review and act.

    ProjectbetaTypeScript
  • jevmory

    Local memory for coding agents that stores verbatim quotes and has Jev grade memory lines as keep, stale, wrong or unsupported.

    ProjectbetaPython
  • jev-agent-kit

    Typed judgment layer for Claude Code, Codex and Hermes Agent: deterministic facts in code, language judgments in Jev, with receipts.

    ProjectbetaPython
  • jev-carryforward

    MCP server with record and recall tools: facts are logged during a session and Jev keeps only those relevant to the next task.

    MCP serverbetaTypeScript
  • hippo-memory

    Local long-term memory for coding agents with an optional Jev reranker that orders recalled memories.

    ProjectproductionTypeScript
  • jev-mcp

    MCP server giving any agent Jev's classify, score, yes/no check and allow/confirm/block gating of risky tool calls.

    MCP serverbetaJavaScript