Every morning, same thing.
Open Claude Code. Start typing. Claude asks me to explain the project architecture. Again.
The decisions we debated yesterday? Gone. The bug we spent 2 hours debugging? Gone. The reason we chose SQLite over PostgreSQL? Gone.
Groundhog Day. Every single session.
I tried everything.
Notion — I never look at it. Claude never looks at it. Dead memory.
Pasting context manually — 5000 tokens of project brief every session. Not scalable.
Claude Projects — doesn't persist decisions across sessions.
None of it works because the AI can't query it, can't learn from it.
So I built Pandaibesy.
It's a Python CLI. 100% offline. No pip install. Pure Python standard library only.
Three commands:
Save a decision during your session
python nalar_bridge.py capture "We use SQLite, not PostgreSQL" myproject
Before starting work, pull relevant context
python nalar_bridge.py query "database decision"
Inject into your agent session
python nalar_bridge.py mcp-pull "working on database layer"
The mcp-pull output is a clean XML block that Claude Code reads directly:
[score:0.50 | source:capture | project:myproject]
We use SQLite, not PostgreSQL
No cloud. No API keys. No subscription. Data stays on your machine forever.
Works on Termux (Android) too — tested on Python 3.13.
Setup:
git clone https://github.com/pandaibesy/pandaibesy.git
cd pandaibesy
python nalar_bridge.py
That's it. Under 3 minutes.
I'm looking for 3-day crash testers — people who use Claude Code or Aider daily.
After 3 days: run python nalar_bridge.py export-logs, paste the JSON into a GitHub Issue. That's the whole ask.
Repo:
https://github.com/pandaibesy/pandaibesy
If you've been living this Groundhog Day problem — I'd love to know what breaks.
Top comments (0)