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

fast-jev-compaction

Fork of fast-jev-compaction adding env-configurable endpoints, a CLI, an MCP server and a Devin post-compaction restore hook.

LibrarybetaMIT
View on GitHub ↗

Overview

This is a modified fork of tamaratran/fast-jev-compaction. The core remains the same: a Claude Code plugin and npm library that replaces the compaction summary with Jev decisions, scoring every tool call and result in one fast request, dropping or truncating stale ones and keeping everything else verbatim, with the conversation state fitted to a token budget in stages.

The fork adds JEV_* and TYPESAFE_* environment variable resolution so it can target any System One-compatible endpoint, a command-line interface, a stdio MCP server (jev-compaction-mcp), and a Devin PostCompaction restore hook that re-injects verbatim tool details lost by Devin's built-in summary. As upstream, calls in the first message and the newest messages are pinned and never touched, and the state sent to Jev replaces each tool result with a short note.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this tool call or result still needed?(inferred)yes · no

How it works

  1. Configure a System One-compatible endpoint via env vars
  2. Score tool calls and results with Jev
  3. Keep, truncate or drop items
  4. Use via Claude Code, CLI, MCP or the Devin restore hook

Strengths

  • Works with any compatible endpoint
  • CLI and MCP access
  • Adds a Devin integration

Caveats

  • Fork; changes may diverge from upstream

Getting started

Install the package or plugin and set JEV_*/TYPESAFE_* environment variables; run jev-compaction-mcp for MCP use.

At a glance

Type
Library
Status
beta
Language
TypeScript
Runtime
npm library, CLI, MCP server, Claude Code plugin
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does this fork add?

Configurable endpoints, a CLI, an MCP server and a Devin post-compaction restore hook.

Is the compaction logic different?

No, it follows upstream.

Can it use a local model?

Yes, any System One-compatible endpoint can be configured.

Related projects

  • 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
  • jev-rules

    Claude Code plugin where Jev asks one yes/no per standing rule and delivers only the rules relevant to each prompt, once per session.

    ProjectbetaJavaScript
  • jev-skill-gate

    Scores installed Claude Code skills against the current project with Jev and hides irrelevant ones via skillOverrides.

    Agent skillbetaJavaScript
  • jev-skill-selection

    Pre-message hook that keeps or drops SKILL.md files before the first model call, using local overlap or Jev keep/drop judgments.

    HookbetaPython
  • jev-skills

    Claude Code and Codex plugin that removes skills from context and lets Jev decide each turn which SKILL.md bodies to inject.

    PluginbetaTypeScript
  • JevScout

    Toolkit where Jev condenses large MCP responses for Claude Code and maps Codex requests to reviewed, fixed repository operations.

    ProjectbetaTypeScript