User repository on GitHub
基于项目逻辑的dsh记忆系统，项目基于交接文档，知识基于项目沉淀。以人实际工作的逻辑，做知识的聚合和交接。开放书写规范以及全插件提示词的前端编辑（热加载），用户可自行优化，或者让模型帮插件通过修改提示词自进化。
- Forks
- 1
- Last pushed
- Updated Oct 2026
Repo stats from the GitHub API, cached Oct 2026.
Jev directory
CommunityJev Agents & MCP
Dsh Memj
A memory plugin for the DeepSeek Harness that binds knowledge and handoff documents to projects in one local Markdown directory. Its settings include a decision-model page for JEV-style models over the /v1/systemone protocol, used for active injection and end-of-turn scoring.
- Stars
- 1
- Language
- JavaScript
- License
- MIT
- Reading time
- 2 min
- Published by
- Community
- Author
- jojoman2024
- Last verified
- Oct 2026
- Languages
- JavaScript
- Capabilities
- Open sourceRuns locally
Highlights
- Projects are the memory entity: one editable project document is the entry point across sessions and instances.
- Settings include a decision-model page for JEV-style models over /v1/systemone, separate from OpenAI-compatible models.
- Two Jev-driven paths: active knowledge injection, marked experimental, and end-of-turn hot-knowledge scoring.
- Library is one directory, ~/.dsh-memj/, in Markdown: copy it to migrate, diff it, edit it anywhere.
- Knowledge types, document templates and all prompts are editable in the UI with hot reload, no code changes.
Quickstart
bash
dsh plugin --profile web add dsh-memjWatch out
README and GitHub metadata reviewed 2026-10-17; the digest truncates the knowledge-management sections, and the JEV decision-model integration is documented deeper in the full README capture than in the excerpt. The author flags active injection as experimental.
Questions about this entry
Short answers drawn from the linked project or page.
A memory plugin for the DeepSeek Harness that treats each project as the entity its knowledge belongs to. Handoff documents and accumulated knowledge live in one local directory of Markdown files and travel with the project.
Its settings include a dedicated decision-model configuration page for JEV-style models. Those models use a different protocol from OpenAI-compatible endpoints, sending model, state and questions to /v1/systemone and reading answers from the response.
Two documented paths: active injection, where a sidecar judges whether the current step needs knowledge the model did not fetch and injects it, and end-of-turn scoring of hot knowledge. The author flags active injection as experimental.
Recommended: open the Plugins page in DSH, click Add plugin, type dsh-memj and install, with no commands or restarts. The equivalent command line is dsh plugin --profile web add dsh-memj; a DSH version requirement applies.
Because JEV-style decision models are not OpenAI-compatible: they speak /v1/systemone, so the harness's native model page cannot hold them. The plugin's page supports multiple providers with independent, write-only keys.
Knowledge types, document templates and every prompt are editable in the frontend with hot reload. The author's example: you can make the model fork a sub-agent to persist knowledge before finishing by changing the tool protocol and prompts.
The whole library is one directory, ~/.dsh-memj/, and project documents and knowledge are Markdown files. Migration is copying the directory, and everything stays readable, diffable and editable outside the tool.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: jevmem keeps Claude Code project memory by asking Jev which turns are worth saving.
avinashjetwani61 pointsJevmem – automatic project memory for Claude Code, built on JevRead the thread on Hacker NewsMore Jev resources like this
An opencode security plugin powered by Jev that risk-scores every tool call before it runs and scans every result for prompt injection afterwards. Deny blocks, ask prompts and allow runs, with configurable thresholds and an optional fail-closed mode.
Jev Agents & MCP#community#github#javascript
Community
surfaceflinger/jew-guardJavaScript
surfaceflinger/jew-guardJavaScriptA very WIP tool for jev-ification from Harry Solovay: lessons from coding sessions become approved Markdown traits, and a configurable decision model evaluates code one trait at a time while an MCP server handles group-aware approval.
Jev Agents & MCP#community#github#typescript
Community
harrysolovay/traits-pocTypeScript
harrysolovay/traits-pocTypeScriptLocal-first MCP memory server for IDEs: tree-sitter indexing into a local ChromaDB store, plus an optional Jev layer that screens retrieved passages for relevance, evidence, contradiction and prompt injection.
Jev Agents & MCP#community#github#python
Community
KikeVen/zerikai_memoryPython
KikeVen/zerikai_memoryPythonAn interactive site by Sarah Drasner that explains System One thinking — fast, automatic judgments versus deliberate checking — and what that distinction means for software design, closing with the Jev × WebMCP Chrome extension as a case study.
Jev Guides & Articles#community#github#javascript
Community
sdras/system-one-explainerJavaScript
sdras/system-one-explainerJavaScriptDjaevin, a calibrated, local-first decision layer from Icaro0310 that runs behind a Devin ACP backend with no API key. It ships as an MCP server published in the official MCP Registry and installs from PyPI as poordjaevin; the repo topics name calibration, conformal prediction and NLI.
Jev Agents & MCP#community#github#python
Community
icaro0310/devin-judgePython
icaro0310/devin-judgePythonA skill for Codex agents that splits development work into Bounded Tasks and produces a Dispatch Plan recommending a model, reasoning effort and tools per task, with optional Jev-powered assessment and an Agent fallback when Jev is unavailable or low-confidence. MIT-licensed.
Jev Agents & MCP#community#github#javascript
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.