I was quite pissed off of Anthropic native UI for checking usage limits, it seemed to be too minimal. I search for various OS projects but they didn’t quite match my idea, I was interested in a all-round control panel for Claude Code which was also a single binary with no dependencies or installation steps. The other point that caught my attention was the forecasting of the usage limits: in some other popular projects like CCUM seemed too simplistic and I wanted a calibrated statistical model for the prediction, with related credible intervals.
I built claumon in the last few months using Go, supporting Linux, MacOS and Windows and also brew installation. I added the usual consumption gauges, cost breakdowns, conversation history but also 2 tabs dedicated to memory management, since after a while I had different memories scattered around several projects and I wanted an intuitive way to analyze those and eventually remove stale ones.
In the last few weeks I worked on the usage forecast model, starting from an empirical Bayes linear regression with added Brownian noise, but finally landing on a Gamma process for the path noise, since token usage cannot decrease in time, and Brownian modeling, even if mathematically convenient, was not the right choice. I added the calibration part based on the recorded historical data, there is a dedicated tooling for that.
I wrote a full formal versioned spec for the model, which is then used for the implementation:
https://github.com/fabioconcina/claumon/blob/main/internal/f...Install is easy:
brew install fabioconcina/claumon/claumon
or use a prebuilt binary from releases.
Everything is local, no data leaves the machine, license is MIT.
https://github.com/fabioconcina/claumon
I would really like to receive some feedback, especially on the statistical model.
Thanks!