Claumon – Forecasting Claude Code usage limits with a Gamma process

Claumon, a new open-source dashboard tool for Claude Code, provides real-time rate limit gauges, token and cost breakdowns, and 14-day usage trends via a single binary with no dependencies. The tool, available for macOS, Linux, and Windows, offers session forecasts with 80% confidence intervals, per-session token analysis, and a memory browser with relationship graphs and staleness alerts. Users can install it via Homebrew, download prebuilt binaries, or build from source, with optional service installation for automatic startup on login.

Claude Code dashboard - run it, see where your tokens go. Single binary, zero config, one browser tab. Runs on macOS, Linux, and Windows. There are plenty of Claude Code dashboards out there. This one isn't trying to be special - it's just my take on it, optimized for the things I care about: minimal, fast, and easy to set up. Single binary, no dependencies, no build step, no config required. Run it and open a browser tab. It gives you rate limit gauges with usage forecasts, per-session token breakdowns, cost estimates, historical trends, conversation history, and a memory browser with relationship graph, health scores, and staleness alerts. Everything updates in real time via SSE. Daily aggregates are stored in SQLite so you can track usage over weeks, not just the current session. Live dashboard: rate-limit gauges, token and cost breakdowns, and 14-day trends. Session forecast: projected usage at reset with an 80% confidence interval. Homebrew macOS / Linux : brew install fabioconcina/claumon/claumon claumon Download a prebuilt binary from the latest release https://github.com/fabioconcina/claumon/releases/latest : macOS Apple Silicon curl -Lo claumon https://github.com/fabioconcina/claumon/releases/latest/download/claumon-darwin-arm64 chmod +x claumon ./claumon macOS Intel curl -Lo claumon https://github.com/fabioconcina/claumon/releases/latest/download/claumon-darwin-amd64 chmod +x claumon ./claumon Linux x86 64 curl -Lo claumon https://github.com/fabioconcina/claumon/releases/latest/download/claumon-linux-amd64 chmod +x claumon ./claumon Windows x86 64 Download claumon-windows-amd64.exe from the releases page If blocked by Defender: right-click the .exe - Properties - Unblock Or run: Unblock-File claumon-windows-amd64.exe Build from source: go install github.com/fabioconcina/claumon@latest claumon Open http://localhost:3131 http://localhost:3131 in your browser. claumon reads credentials from ~/.claude/.credentials.json created by claude /login , or falls back to the OS credential store macOS Keychain, Windows Credential Manager used by the VS Code extension. If credentials are missing, session tracking still works. Only the API usage gauges are unavailable. Run claumon automatically on login so the dashboard is always available. First, move the binary to a permanent location, then register the service. macOS: mkdir -p ~/.local/bin mv claumon ~/.local/bin/ Add to PATH if not already: export PATH="$HOME/.local/bin:$PATH" claumon service install Creates a LaunchAgent at ~/Library/LaunchAgents/com.claumon.dashboard.plist . Logs go to ~/.claumon/claumon.log . Linux: mv claumon ~/.local/bin/ claumon service install Creates a systemd user unit at ~/.config/systemd/user/claumon.service . Logs go to journald journalctl --user -u claumon . By default, systemd user services stop when you log out. To keep claumon running after logout, enable lingering: loginctl enable-linger $USER . Windows PowerShell : Move the exe to a permanent location New-Item -ItemType Directory -Force "$env:LOCALAPPDATA\Programs\claumon" Move-Item claumon-windows-amd64.exe "$env:LOCALAPPDATA\Programs\claumon\claumon.exe" Add to PATH current user, persists across sessions $p = Environment ::GetEnvironmentVariable 'Path', 'User' Environment ::SetEnvironmentVariable 'Path', "$p;$env:LOCALAPPDATA\Programs\claumon", 'User' Restart your terminal, then: claumon service install Adds claumon to the Windows Startup folder - it launches automatically hidden on login. Manage the service: claumon service status check if running claumon service restart restart after config changes claumon service uninstall stop and remove No root or admin required on any platform. claumon update Checks GitHub releases for a newer version, downloads the right binary for your platform, and replaces the current one. If a background service is installed, it restarts automatically after the update. Remove the background service if installed , delete the binary, and optionally remove local data SQLite database, logs, config, pricing cache . macOS / Linux: claumon service uninstall rm ~/.local/bin/claumon rm -rf ~/.claumon Windows PowerShell : claumon service uninstall Remove-Item Get-Command claumon .Source Remove-Item -Recurse "$env:USERPROFILE\.claumon" Session usage - 5-hour sliding window utilization with reset countdown Weekly usage - 7-day window utilization with reset countdown Per-model quotas - separate Opus and Sonnet weekly limits when applicable Extra usage credits - monthly limit and spend if enabled Gauges are color-coded: green <50% , yellow 50–80% , red 80% . Projected utilization at reset - an 80% credible interval and ETA to threshold for each rate-limit window, so you can see where you'll land before you get there. Empirical-Bayes model, refit daily from your own past windows, over a monotone non-decreasing usage process so the interval respects that utilization only grows within a window. Full spec in LaTeX source: internal/forecast/MODEL.pdf . MODEL.tex Benchmarked out-of-sample with claumon bench - leave-one-out and temporal-holdout protocols, proper scoring CRPS/pinball with coverage and bias breakdowns. Input / output / cache read / cache write tokens per session Estimated equivalent API cost using current model pricing Daily aggregates with 14-day trend charts Activity heatmap - 24-hour breakdown of token activity by hour Active sessions table - project, model, tokens, cost, messages, last activity Today / Recent toggle - switch between today's sessions and the 50 most recent across all time Running process detection - shows which sessions are actively running with green "active" badge Session detail view - full message timeline with per-message token counts and tool usage Automatic discovery - watches ~/.claude/projects/ for new and updated sessions Live process table - shows all running Claude Code processes with PID, chat title, project, type, and uptime Stop button - send SIGINT to gracefully stop any running process conversation is preserved on disk Auto-refresh - updates via SSE and periodic polling All memory files - CLAUDE.md, rules, auto-memory indexes, per-project memories Search across content, path, project name, and frontmatter Filter by category - CLAUDE.md, Rules, Index, Memory Staleness indicators - green today , gray <7d , yellow 7–30d , red 30d Staleness alerts - broken MEMORY.md links, orphaned files, index mismatches VS Code integration - click to open files in your editor via vscode:// links Per-file grading - each memory file gets a letter grade A–F based on freshness, structure, specificity, and connectedness Improvement suggestions - actionable tips for low-scoring files add frontmatter, link from MEMORY.md, etc. Interactive visualization - nodes are memory files, edges show relationships Project filters - focus on specific projects Clickable legend - toggle node types on/off Click to navigate - click a node to jump to its file in the memory browser | Key | Action | |---|---| 1 2 3 | Switch to Dashboard, Memories, Graph tab | / | Jump to memory search | Esc | Close session detail panel | The dashboard uses Server-Sent Events SSE - no polling, no manual refresh. Changes appear instantly: Usage gauges update every 2 minutes configurable from the Claude API Session table updates when any .jsonl session file changes on disk Memory browser highlights changed files with a visual pulse Graph and staleness re-render when memory files change on disk A status dot in the top bar shows connection state green = connected, red = disconnected . Optional. Create ~/.claumon/config.json : { "port": 3131, "poll interval seconds": 120, "credentials path": "~/.claude/.credentials.json", "claude dir": "~/.claude", "db path": "~/.claumon/usage.db", "stuck threshold minutes": 10 } All fields are optional. Defaults are shown above; claumon works without a config file. claumon stores usage snapshots and daily aggregates in a SQLite database at ~/.claumon/usage.db . Historical data is backfilled automatically on first startup by scanning all existing session files. The database uses WAL mode for concurrent reads during writes. No maintenance required. claumon combines three data sources: Claude API /api/oauth/usage - polled periodically for rate limit utilization. Requires OAuth credentials from claude /login . Session files ~/.claude/projects/ / .jsonl - watched via fsnotify for real-time token counting. Each JSONL file is parsed for message-level token usage. Memory files ~/.claude/projects/ /memory/ .md , CLAUDE.md , rules - discovered and parsed for frontmatter, rendered as HTML, analyzed for cross-references and staleness. Everything is embedded in a single binary via //go:embed - no external files, no Node.js, no build step for the frontend. php graph TD API "Claude OAuth API" -- ENGINE JSONL "Session JSONL Files" -- ENGINE MEM "Memory Files" -- ENGINE PRICE "Pricing Data" -- ENGINE ENGINE "Pollers · Watchers · Parsers" -- DB "SQLite" ENGINE -- CACHE "In-Memory Cache" ENGINE -- SSE "SSE Broker" DB -- HTTP "HTTP Server" CACHE -- HTTP JSONL -- |"parsed fresh"| HTTP SSE -- HTTP HTTP -- SPA "Embedded SPA" MIT