Build local shared memory when you work across projects and AI tools.
Last reviewed:
About this guide
This guide was prepared with AI assistance using the project’s public documentation. It does not claim testing on every device or independent human review. Project publisher: AmirSina Zamanian.
Sources are listed at the end. Use the project repository to report an error or suggest a correction. Report an issue
⌕
Search runs on this device.
On this page
Overview
This system is for developers who move between Codex, Claude Code, or Gemini CLI and want projects and approved decisions in one local memory. If you need automatic cloud sharing of private content, this local-first model is not the right fit.
What problem does it solve?
When several tools work on one project, decisions and work status can become scattered. The system maintains approved working memory in a shared local layer; it does not replace coding tools or provide a cloud collaboration service.
Data and permissions
You select the sources. Local storage does not make every file safe: review exports before transfer and do not publish credentials, private text, or sensitive paths. Information without evidence must not become a verified fact.
What to expect from measurements
actual uses verifiable data; estimated is a forecast; unavailable means required data or pricing is missing. Costs use a confirmed price book. Missing values are not zero, and a cost comparison alone does not establish savings caused by the system.
Completion checklist
Before continuing, identify the tools you use, which projects are genuinely shared, and which information may be transferred. If you only need measurement views, read the Dashboard-only limitations. For shared memory, continue with environment preparation and the Obsidian decision.
Decide before installing
For a single short question, a memory system may be unnecessary. Explore this workflow when multiple sessions or tools work on a continuing project and decisions get lost. Start with one small project and define success first: the next tool should accurately restate the decision and next action. Do not assume cost reduction or complete memory as a default product outcome.
The related public lab uses synthetic data without real usage sources. It demonstrates adapter and context-pack creation within a defined scenario only. For a final decision, evaluate continuation quality in your own approved project; a successful script is not a substitute for evaluating the receiving tool’s response.