Agent monitoring
Blume
Desktop sidecar that monitors and improves your coding agents
Blume is a local, private desktop sidecar that watches your coding agents (Claude Code, Cursor, Codex, Pi, and more). It shows what every agent is doing in real time, tracks the hidden rules, skills, hooks, and instructions that shape their work, and surfaces concrete suggestions to keep your agent setup consistent - e.g. spotting when a workflow was re-explained three times and proposing it become a skill. Conversation history stays on-device.
What Blume is
Blume is a quiet window into every coding agent you run. It sits as a narrow desktop sidecar, watches your harnesses (Claude Code, Cursor, Codex, Pi, omp), and tracks the hidden files behind them - rules, skills, hooks, and instructions - so you can see what is actually guiding each agent instead of guessing.
What it does
- Agent overview - see what each coding agent is doing right now and the exact moment one finishes or needs your approval.
- Hidden files and rules - track the rules, skills, hooks, and agent config files that quietly shape what agents do, across Claude Code, Cursor, Codex, and more, all in one place.
- Usage tracking - see how much of each provider plan you have left before you hit a wall.
- Improve - Blume reviews your rules, skills, and hooks on-device and proposes fixes; preview, then approve or dismiss.
- Private by default - conversation history is stored locally on your device.
Honest caveat
Blume is an observability and improvement layer, not a coding agent itself. Its value comes from acting on its suggestions (adopting skills, tightening rules), so it is most useful once you have a real agent setup to monitor.
At a glance
Who's behind it
Blume
- Source
- closed-source
- Platforms
- macos, windows, linux, desktop
- Install
Download the desktop app from blume.codes- Model support
- Works alongside Claude Code, Cursor, Codex, Pi, and omp - it is not itself a model provider
- Website
- https://blume.codes