Sabbir Amin
In my last post I mentioned we were experimenting with a "global context archiver" for our agentic workflows. I spent the previous weekends building around this concept. It is called MemoryBase 📝 (github.com/sabbiramin113008/memorybase).
Nothing fancy, just a small tool I put together to solve a problem that was genuinely frustrating me.
The problem: every time a new agent session starts, context is lost. Decisions made two days ago are gone. Tasks get repeated. The agents drift. So I built three simple primitives around it:
Blueprint: your project definition, stack, folder structure, constraints.
Tasks: a Kanban board agents can read and update in real time.
Vault: a searchable log of decisions, errors, and notes across sessions.
All of it is accessible via 17 MCP tools. Works with Claude Code, Cursor, or any MCP-compatible client. SQLite out of the box, PostgreSQL if you need it in production.
Getting started is just: pip install memorybase
That's it to start and play around. Find more details in the linked GitHub Pages.
GitHub:
github.com/sabbiramin113008/memorybase
PyPI:
pypi.org/project/memorybase
Docs:
sabbiramin113008.github.io/memorybase
It is early, rough around the edges, and very much a work in progress. But it has already made my own agentic workflows a lot more stable. If you are building with agents and context loss is a problem for you too, give it a try. Feedback welcome, good or bad. 😄