jevmemjevmem

How do I install jevmem?

Install jevmem in three steps: add the plugin in the Claude app (Plugins → Discover → jevmem → Add), install the CLI with npm install -g jevmem, and run jevmem setup in your project's folder, which asks for your TypeSafe API key without showing it, turns jevmem on there and checks the setup. If you start Claude Code before that is done, the plugin says once what is left.

Last updated: · jevmem 0.7.8 · Install · Source on GitHub

Once it runs, deciding what to save takes 0.16 s and costs $0.00019 per message, at TypeSafe's listed price for Jev (measured on 66 held-out turns with jevmem 0.7.7, one run on 2026-11-08, results). The rest of this page is the repository's install and upgrade notes.

You need a TypeSafe AI key for Jev. jevmem writes each line itself; an OpenAI or Anthropic writer is optional and off unless you set writer in jevmem.config.json (configuration).

Option 1: Claude Code plugin (recommended)

From the Claude plugin directory #

This needs Claude Code 2.1.273 or later: earlier versions don't sync the plugins you add in the Claude app (Claude Code docs).

  1. In the Claude app: Plugins → Discover → jevmem → Add. The app warns you before it adds the plugin; the warning is about the plugin's local MCP server, jevmem mcp, a command the plugin runs on your computer (the CLI from step 2).

  2. Install the CLI the plugin runs:

    npm install -g jevmem
    
  3. In a terminal, in your project's folder, finish the setup in one go:

    jevmem setup
    

    It says which jevmem and which plugin it found; asks for your TypeSafe key (from console.typesafe.ai/keys) without showing it as you paste, and saves it in ~/.jevmem/env, readable only by you (a key already saved is kept); turns jevmem on in that folder, where the plugin does nothing until then; checks the key with one request to TypeSafe (the fixed words "jevmem setup: key check", nothing from your project: PRIVACY.md); runs jevmem doctor; and, when the project has past Claude Code sessions, offers to import them: the count and the cost first, then the lines it would add, and nothing is sent or written before you say yes to each. By hand instead: jevmem key, then jevmem enable in the folder.

  4. Start Claude Code in that project, signed in with the same Claude account as the app. The plugin shows as jevmem@synced (run /reload-plugins if Claude Code asks).

  5. jevmem doctor checks the setup any time. If Claude Code shows "jevmem: CLI not found", see if the plugin can't find the CLI.

If you start Claude Code before the setup is done (since 0.7.2), the plugin says so once, at the start of a session, and names what is left: the jevmem command, or this project not turned on yet, with jevmem setup as the one command, which asks for your key if none is saved. It is shown once per computer, in a terminal, the desktop app's Code tab, the VS Code extension or an SSH session of the desktop app, and never again after it has been shown, after the setup is done, or after you opt out; a cloud or Cowork session gets no message and no marker (since 0.7.3; 0.7.2 showed it at the start of every cloud chat); it sends nothing and writes nothing inside the project (the one-time setup message). A terminal shows it as a line from the SessionStart hook; in the Claude desktop app's Code tab, which does not draw such lines, Claude passes it on at the start of its first reply.

Claude Code's own memory may also say it saved something; JEVMEM.md shows what jevmem saved.

From the jevmem marketplace #

npm install -g jevmem
claude plugin marketplace add Avinash-jetwani/jevmem
claude plugin install jevmem@jevmem
cd your-project && jevmem setup

jevmem setup asks for your key (or run jevmem key); for a plugin installed from a marketplace you can instead enter it in Claude Code with /plugin configure jevmem, which Claude Code keeps in your system's secure credential store (the claude plugin install shell command doesn't ask for it). That setting exists only for a plugin installed from a marketplace: the directory's jevmem@synced has no Configure options.

The side pane (0.7.5) #

A second plugin from the same marketplace draws a pane beside your Claude Code chat with what jevmem recalled for the prompt, what the guard decided and what the turn saved (docs/pane.md):

claude plugin marketplace add Avinash-jetwani/jevmem   # once, if not added yet
/plugin install jevmem-pane@jevmem                      # in Claude Code

It needs the jevmem command (0.7.5 or newer), a project that is turned on, and Claude Code 2.1.299 or newer; the jevmem plugin itself is unchanged.

The Cursor plugin (0.7.6) #

jevmem is a Cursor plugin too: cursor-plugin/ in this repository, listed in the repository's .cursor-plugin/marketplace.json as jevmem. It carries the five hooks that save each finished agent turn, the MCP server the agent reads memory over, and the rule that tells it to, in every Cursor window, so a project needs no .cursor/ files from jevmem. Like the Claude Code plugin it runs the jevmem CLI from npm through one shell launcher and ships no code; it needs the CLI and a TypeSafe key saved with jevmem key, and it does nothing in a project until you turn jevmem on there.

  1. Install the CLI and save your key, as above: npm install -g jevmem, then jevmem key.
  2. Install the plugin. From the Cursor Marketplace once jevmem is listed there (open Customize in Cursor, find jevmem, Install). Until then, from the repository: copy cursor-plugin/ to ~/.cursor/plugins/local/jevmem and run Developer: Reload Window; Cursor loads plugins from that folder. jevmem's listing on cursor.directory (since 2026-10-23) is not the plugin: its Add to Cursor button installs the MCP server alone, with the entry in docs/mcp.md, so the agent can read memory there; saving needs the plugin folder or jevmem init --tool cursor.
  3. In each project: jevmem setup (asks for the key if none is saved, turns jevmem on there, checks the setup) or jevmem enable. The plugin's MCP server runs in every window from the start, with no toggle (Cursor starts a plugin's server on load, unlike a project's .cursor/mcp.json server); in a project not turned on it answers that jevmem is off there, and the hooks and the rule do nothing.

A project set up earlier with jevmem init --tool cursor keeps working beside the plugin, which stands down there so nothing is saved twice; jevmem init --remove-hooks moves it over to the plugin alone (it also removes jevmem's rule file and MCP entry, which the plugin carries). What the plugin runs, what Cursor does with it, and what was measured: docs/hooks.md.

If the plugin can't find the CLI #

The plugin runs the jevmem CLI from npm, so install that first. The hooks find it on the PATH Claude Code gives them or, when that PATH lacks it (the desktop app's can), in /opt/homebrew/bin, /usr/local/bin, ~/.local/bin, ~/.volta/bin or the newest Node version under ~/.nvm. Without it, an enabled project shows "jevmem: CLI not found, so memory is off in this project" on the first prompt of each session, and the MCP server fails to start (/mcp shows it as failed). The plugin does nothing until you run jevmem enable in a project; what it runs, and how to switch it off: docs/hooks.md.

Option 2: npm (also sets up Cursor and Codex; for Cursor, jevmem init --tool cursor is the whole setup on disk: the hooks, the launcher, the MCP entry and the rule; one toggle in the Cursor app then turns the project's MCP server on, the step. With the Cursor plugin installed, init only turns the project on)

npm install -g jevmem
cd your-project
jevmem init --tool claude

init creates JEVMEM.md, jevmem.config.json and .jevmem/, and registers the four Claude Code hooks: Stop, UserPromptSubmit, PreToolUse and, since 0.7.1, SessionStart (details). Hooks don't get your shell's variables and jevmem doesn't read shell profiles, so save the key with jevmem key, which puts it in ~/.jevmem/env. jevmem doctor checks the setup.

Already have a CLAUDE.md? jevmem import splits CLAUDE.md, AGENTS.md and .cursor/rules/* into statements, puts each through the same gate as a turn, and prints what it would add; --apply writes them. --from claude-auto-memory also reads Claude Code's own auto memory for the project. The source files are only read. Since 0.7.1, jevmem import --from sessions reads the project's past Claude Code sessions the same way, after showing the count and the cost and asking (how it works).

Works with #

What is automatic and what depends on the agent:

Tool Setup Capture Recall
Claude Code the plugin, or jevmem init --tool claude Automatic, every turn, via the Stop hook Automatic, every prompt, via UserPromptSubmit
Codex jevmem init --tool codex Automatic while jevmem watch runs (it tails Codex's session log for this project and runs the same decide → write path); otherwise agent-initiated via MCP add_memory, prompted by an AGENTS.md section Agent-initiated: search_memory via MCP, prompted by AGENTS.md
Cursor the Cursor plugin (0.7.6), or jevmem init --tool cursor Automatic since 0.7.4, every finished agent turn, via Cursor's hooks (the plugin's, or the project's in .cursor/hooks.json; the app's stop hook, the agent CLI's sessionEnd), through the same decide and write path as Claude Code (what was measured). Before 0.7.4: only when the agent called MCP add_memory Agent-initiated: the rule (the plugin's, or .cursor/rules/jevmem.mdc) tells it to call search_memory before non-trivial tasks (Cursor's prompt hook takes no context, so recall is not automatic there). The plugin's MCP server runs in every window from the start; a project's .cursor/mcp.json server starts only after you turn it on once in the Cursor app, Customize → MCPs → jevmem → Configure → the source toggle for .cursor/mcp.json, then Reload (the step)
Claude Desktop jevmem init --tool claude-desktop prints a config snippet to paste (one project per config, named with --root) Manual: ask it to call add_memory (no hook, no rule file) On request: search_memory

MCP add_memory goes through the same gate as the hook. Client configs: docs/mcp.md.

Upgrading #

Upgrading to 0.7.2. npm install -g jevmem@latest; the plugin's SessionStart entry gains a flag (--start), so the directory and marketplace plugins move to 0.7.2 with the CLI (a 0.7.2 plugin with a 0.7.1 CLI prints the usual "older than this plugin" line; its one-time setup message needs no CLI at all). What changes for you: an always-on line is served at session start only when it is verified or the poisoning gate has passed it, so a line you marked on 0.7.1 by hand on a line jevmem did not write waits for its verdict until your next prompt; jevmem always <id> and jevmem add --always now ask the gate once (PRIVACY.md). The default Jev model is pinned to jev-1.13.0; a jevmem.config.json written before 0.7.2 says "model": "jev-latest" and follows TypeSafe's alias (to 1.14.0 from about 2026-10-21): jevmem doctor says so, and "jev": {"model": "jev-1.13.0"} pins it (configuration).

Upgrading to 0.7.1. npm install -g jevmem@latest, and the plugin gains a fourth hook, SessionStart: the directory and marketplace plugins move to 0.7.1 with the CLI (a 0.7.1 plugin with a 0.7.0 CLI prints the usual "older than this plugin" line, and its SessionStart hook runs a CLI that ignores the event); with jevmem init hooks, run jevmem init --tool claude again in each project to add the hook (nothing else changes). A jevmem.config.json from before has no alwaysOn and no thresholds.alwaysMin and gets the defaults (always, 0.5). Lines saved before 0.7.1 carry no always-on mark: jevmem always --suggest proposes marks for the live rule-like lines and --apply sets them, or mark by hand with jevmem always <id>. A line marked always-on (always:1 in its comment) is read as usual by a 0.7.0 teammate, whose jevmem drops the mark when it rewrites the file. One more question leaves the machine in the request each turn already sends (asked in each tier), no new text (PRIVACY.md); jevmem import --from sessions reads past sessions only when you run it and say yes.

Upgrading to 0.7.0. From 0.6.0 to 0.6.6, npm install -g jevmem@latest is the whole upgrade: the plugin's hooks have not changed since 0.6.0, and the directory and marketplace plugins move to 0.7.0 with the CLI. 0.7.0 asks one more question in the request each turn already sends (PRIVACY.md); a jevmem.config.json written earlier has no thresholds.duplicateMin and gets 0.7. jevmem wrong … --should-be none now retires the line instead of deleting it, and a [retired] line stays in JEVMEM.md: a teammate still on 0.6.x keeps it as text, and moves it after the memory lines when their jevmem rewrites the file. Nothing else changes in what is saved, recalled or checked.

Upgrading to 0.6.6. From 0.6.0 to 0.6.5, npm install -g jevmem@latest is the whole upgrade: the plugin's hooks have not changed since 0.6.0, and the directory and marketplace plugins move to 0.6.6 with the CLI. 0.6.6 reads Claude's reply on more turns (PRIVACY.md); "jev": { "askAboutReply": false } in jevmem.config.json keeps 0.6.5's behaviour. From 0.5.x, one of the paths below, tested on 2026-09-30 on the 0.6.0 build in a temporary HOME with a project set up on 0.5.10 that kept its lines (results/upgrade-2026-09-30.txt); 0.6.1 to 0.6.6 change nothing in them:

The limits, in short #

Every limit, with the numbers: the FAQ.