I Built Open-Source AI Dev Tools to Close the Gap Between Seeing a Bug and Fixing It

#webdev #ai #javascript #opensource

I've been using AI assistants like GitHub Copilot and Claude for debugging for over a year now. And I kept running into the same frustrating moment: I'd see a bug on my screen, but by the time I described it to the AI, I'd lose half the context — the exact error, what I clicked, what the logs said.

So I built Machina — a suite of open-source tools designed to close that gap.

The Core Problem

When you're debugging with an AI assistant, you need to transfer context: what you see, what happened, what the environment looks like. Doing this manually is slow and lossy. You forget things. You mis-describe things. The AI works with incomplete information.

Machina automates that context transfer.

The Tools

BugCapture

Records your screen + audio, transcribes your voice with Whisper, takes a screenshot, and generates a ready-to-paste .md file with everything an AI needs to understand your bug. No more copy-pasting error messages or describing what you see.

ContextForge

Before starting an AI debugging session, run ContextForge. It pulls your recent git diff, SSH logs, and any BugCapture output into a single structured briefing. Your AI starts the session already knowing what changed and what broke.

LearnBoard

A UI for LEARNING.md — a persistent memory file for your AI. Instead of re-explaining your codebase conventions every session, LearnBoard lets you manage what the AI should always remember. With stats on how often each lesson is used.

PromptBoard

A drag-and-drop canvas for building structured prompts. Combine context blocks, templates, and voice input (via browser or local Whisper) into prompts that consistently get better AI responses.

Why Open Source

I wanted these tools to be something the community can extend and adapt. Every tool is self-contained: a server.js, an index.html, and a package.json. Run bash setup.sh and you're done.

The repo is at github.com/machina-tools/machina.