The latest articles on DEV Community by pandaibesy (@pandaibesy). https://dev.to/pandaibesy https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F3953403%2Fd0db73fd-47fa-40a0-a8d8-71762d87a4f6.png https://dev.to/pandaibesy en pandaibesy Wed, 27 May 2026 02:01:36 +0000 https://dev.to/pandaibesy/i-got-tired-of-claude-code-asking-me-to-explain-my-project-architecture-every-morning-so-i-built-3cdg https://dev.to/pandaibesy/i-got-tired-of-claude-code-asking-me-to-explain-my-project-architecture-every-morning-so-i-built-3cdg <p>Every morning, same thing.<br> Open Claude Code. Start typing. Claude asks me to explain the project architecture. Again.<br> The decisions we debated yesterday? Gone. The bug we spent 2 hours debugging? Gone. The reason we chose SQLite over PostgreSQL? Gone.<br> Groundhog Day. Every single session.<br> I tried everything.<br> Notion — I never look at it. Claude never looks at it. Dead memory.<br> Pasting context manually — 5000 tokens of project brief every session. Not scalable.<br> Claude Projects — doesn't persist decisions across sessions.<br> None of it works because the AI can't query it, can't learn from it.<br> So I built Pandaibesy.<br> It's a Python CLI. 100% offline. No pip install. Pure Python standard library only.<br> Three commands:</p> <h1> Save a decision during your session </h1> <p>python nalar_bridge.py capture "We use SQLite, not PostgreSQL" myproject</p> <h1> Before starting work, pull relevant context </h1> <p>python nalar_bridge.py query "database decision"</p> <h1> Inject into your agent session </h1> <p>python nalar_bridge.py mcp-pull "working on database layer"</p> <p>The mcp-pull output is a clean XML block that Claude Code reads directly:</p> <p><br> [score:0.50 | source:capture | project:myproject]<br> We use SQLite, not PostgreSQL<br> </p> <p>No cloud. No API keys. No subscription. Data stays on your machine forever.<br> Works on Termux (Android) too — tested on Python 3.13.</p> <p>Setup:</p> <p>git clone <a href="https://github.com/pandaibesy/pandaibesy.git" rel="noopener noreferrer">https://github.com/pandaibesy/pandaibesy.git</a><br> cd pandaibesy<br> python nalar_bridge.py</p> <p>That's it. Under 3 minutes.<br> I'm looking for 3-day crash testers — people who use Claude Code or Aider daily.<br> After 3 days: run python nalar_bridge.py export-logs, paste the JSON into a GitHub Issue. That's the whole ask.<br> Repo: <br> <a href="https://github.com/pandaibesy/pandaibesy" rel="noopener noreferrer">https://github.com/pandaibesy/pandaibesy</a></p> <p>If you've been living this Groundhog Day problem — I'd love to know what breaks.</p> ai claude python cli