Blume Watches You Code and Writes the Rules

The frustrating part of working with AI

If you've ever watched someone use an AI coding assistant — or used one yourself — you know the pattern. The AI does something you don't like. You correct it. Next session, same mistake. You correct it again. It never quite remembers the lesson.

Blume fixes that, almost invisibly.

It runs quietly in the background while your developer works with tools like Claude Code or Cursor. It reads the conversation history that's already saved on the computer — no new data sent anywhere — and looks for three things: moments where your dev says "no, not like that," rising frustration (short, terse messages, tasks taking way longer than they should), and gentle reminders that keep coming up ("don't forget to run the tests").

When it spots a pattern repeating, it drafts a rule. Your developer sees the suggestion and approves or rejects it — nothing changes without a human saying yes. Over time, the AI assistant gets quietly, incrementally better at working the way you work.

It also tracks how many AI credits are being used, across different tools, so no one hits a surprise wall mid-project.

Free to download. Runs on Mac, Windows, and Linux. Nothing uploaded to the cloud.

→ Take a look at blume.codes

Words worth knowing

Claude Code / Cursor — AI tools that help write software. Think of them as a very fast junior developer who needs clear instructions.

Session history — The saved transcript of everything your developer and the AI said to each other during a working session.

CLAUDE.md — A simple text file that tells the AI how to behave on a specific project. Blume writes additions to this file automatically.

Token — The unit AI companies use to measure and charge for usage. Like minutes on an old phone plan.