Skip to content

Jevkeep

A Codex plugin that asks Jev which conversation passages still matter, lets Codex compact normally, then restores the selected excerpts verbatim afterwards with labels for their original role and tool source.

Stars
1
Language
TypeScript
License
MIT
Published by
Community
Author
hatt-io
Last verified
Oct 2026

About this project

A Codex plugin that preserves useful conversation excerpts alongside the summary after context compaction. When a conversation gets long, Codex condenses earlier messages into a summary to make room for more work. That summary can lose exact details: a command that failed, a file path, an error message, or a constraint you gave earlier.

Quoted from the project's README. Read the full README.

Highlights

  • It wraps Codex's own compaction: Jev scores relevance before, Codex summarizes as usual, excerpts come back after, so manual and automatic both work.
  • Restored text stays verbatim, each packet is capped at 24,000 UTF-8 bytes, and trimmed excerpts carry an explicit omission marker.
  • Conversation text, tool inputs and textual tool results go to the Typesafe API on jev-latest; reasoning records and binary content are excluded.
  • Missing credentials or a failed scoring pass are non-events: native compaction continues and stale excerpts are discarded, not restored.

Watch out

It preserves selected excerpts, not the whole conversation, and relevance scoring can miss something you need later, at the cost of extra API requests and time on every compaction, with a shared 20-second network deadline inside a 30-second hook timeout. Excerpts belong to one session and transcript, and if Codex's transcript format changes the plugin skips preservation and leaves native compaction running. Setup has sharp edges: Node 22+ must be on the PATH Codex actually uses including desktop launches, hooks must be trusted manually in /hooks since installing the plugin does not enable them, and the key comes from the environment or a 700 directory holding a 600 file that must not be a symlink. Conversation text, which can include source code, leaves the machine on each scored compaction. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

48GitHub stars
Agent harness for work that outlives one chat, with resumable turns, routed state machines, and observational memory across sessions and processes.
Jev Agents & MCP#community#jevbest#catalog-import
Assistant framework that stores specs, memory, and specialist agents as plain markdown files in your repo for Claude Code, Codex, Opencode, and similar runtimes.
Jev Agents & MCP#community#jevbest#catalog-import
36GitHub stars
Obsidian plugin providing multi-provider chat, vault read and write with function calling, visual workflow automation, and local vector search.
Jev Agents & MCP#community#jevbest#catalog-import
36GitHub stars
Pi extension that strips the full Agent Skills catalog from every request and loads only the skills a Jev ranking call judges relevant.
Jev Agents & MCP#community#jevbest#catalog-import
102GitHub stars
Behavioral code search that asks Jev to score repository fragments for a plain-language question and returns exact excerpts with paths and lines.
Jev Agents & MCP#community#jevbest#catalog-import
44GitHub stars
Pi extension that keeps a fast Executor model on implementation and brings in a separate Advisor model for consequential choices and reviews.
Jev Agents & MCP#community#jevbest#catalog-import
Back to all resources

From the Jev community

Posts from builders shipping with Jev right now.

Follow @typesafeai