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

jevmory

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

ProjectbetaMIT
View on GitHub ↗

Overview

jevmory is a zero-dependency memory layer for coding agents. Hooks ingest every session locally into a per-project SQLite store, and remembered facts are verbatim quotes with context rather than generated summaries. Its audit command checks an existing memory file such as MEMORY.md against that evidence and grades each line as keep, stale, wrong or unsupported, with confidence, support and contradiction scores.

There is no LLM generation anywhere: Jev judges and code selects and composes. It installs as a Claude Code plugin with session start and end hooks, offers an MCP server with status, recall and fact tools for Codex CLI, Cursor and other clients, and supports Codex notify hooks. Secrets are redacted before storage and grading only runs after an explicit per-project opt-in. Every judgment and its token usage is stored, so any number in a report traces back to an API call. The README notes that a faithfully stored quote can still be wrong or outdated.

What the decision model decides

TypeQuestionAnswers
ChoiceIs this memory line still valid given the session evidence?keep · stale · wrong · unsupported
Yes / noDoes the evidence support / contradict this claim?—

How it works

  1. Session hooks ingest conversation statements into local SQLite
  2. Opt in to grading for the project
  3. Jev scores support and contradiction for each memory line
  4. Code assigns keep, stale, wrong or unsupported and writes receipts
  5. SessionStart injects top facts into the next session

Strengths

  • Stdlib-only Python, local by default
  • No generated text; facts are verbatim quotes
  • Receipts tie every number to an API call
  • Works with Claude Code, Codex CLI and Cursor

Caveats

  • Grading sends evidence to the hosted Jev API after opt-in
  • Provenance does not establish truth
  • Early status per the README badge

Getting started

In Claude Code: /plugin marketplace add romiluz13/jevmory then /plugin install jevmory@jevmory. Run jevmory audit MEMORY.md to grade a memory file.

At a glance

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

Use cases and integrations

How curated lists describe it

  • Coding-agent memory where every fact is a verbatim quote graded by TypeSafe Jev's calibrated confidence. Local-first, SQLite receipts, zero dependencies.

    logicrw/awesome-jev-projects

  • Coding-agent memory where every fact is a verbatim quote graded by TypeSafe Jev's calibrated confidence. Local-first, SQLite receipts, zero dependencies.

    hellogumbo/awesome-jev

  • Memory for coding agents where every fact is a verbatim quote, graded by Jev into keep, stale, wrong or unsupported, with local SQLite receipts and `jevmory audit` for existing memory files.

    MrJev/awesome-jev

Frequently asked questions

What does jevmory audit decide?

Whether each memory line should be kept or is stale, wrong or unsupported, based on session evidence.

Does any data leave my machine?

Only after you opt in to grading for a project; secrets are redacted at rest.

Does it use an LLM?

No generation is used; Jev judges and code composes the output.

Related projects

  • jevmem

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

    ProjectbetaTypeScript
  • jev_project_context

    Evidence-first experiment-memory skill for coding agents, with an optional Jev audit layer that checks provenance and claim support.

    Agent skillbetaPython
  • 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
  • typesafe-as-a-judge

    Unofficial Codex and Claude Code MCP plugin giving agents bounded Jev tools for routing, ranking, extraction, verification and judging.

    MCP serverbeta
  • hippo-memory

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

    ProjectproductionTypeScript
  • Boson-Video

    Video summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.

    ProjectbetaPython