DBDelve
Dive deep into your data. A fast, native, multi-engine database client written in Rust. Free and open source, no telemetry.
Your Claude Code usage, read from your own machine. No account, no API key, and no network unless you turn projection on.
Developer toolsTypeScript · JavaScriptGitHub @gustav-pjvrOnlineOwner unconfirmed
Have you tried it? One tap is enough. No sign-in needed.
Ambient Windows desktop dashboard for Claude Code and Codex usage: tokens, cost, sessions, models, plan limits. Electron + koffi, no compiler required.
Loading the full README.
Read the full README ↗An ambient Windows desktop dashboard for your AI coding usage, sitting below your normal windows. Go to the desktop and it is there.
Claude Stats is a live command centre for your local Claude Code usage, with OpenAI Codex beside it when Codex is on the machine: tokens, cost at API list prices, hours, sessions, models, projects, plan limits, daily and hourly activity, all read from files on this machine. It has a Simple view (one card each for Claude and Codex) and a Detailed view (every panel, the default), and a switch that hides Codex. It is the only view in the shareable stats edition.
Work (full build only) is the inbox and the monday Tasks board on the wall: mail from Microsoft 365, Gmail and GoDaddy on the left, four scrolling task columns and a summary strip on the right, actions that write back to the mail servers and to monday, and notices. See Work mode.
npm install
npm start # build + launch
Other scripts:
npm run build # bundle to dist/
npm run dev # rebuild on change
npm test # stats, indexing, Codex, plan limits, projection, crypto, chat, pricing, Ask Gemini, Work mode
npm run test:ask-live # Ask Gemini lock-down against the real agy (spends one prompt of quota)
npm run test:render # about 830 layout checks in a real Electron window, every screen size and view
npm run typecheck # tsc --noEmit
npm run package # per-user Windows installer into release/
npm run logo # regenerate the mark, icons and wordmark
npm run deploy # build, upgrade the installed app in place, relaunch
npm run deploy is how the app gets updated. It packages, silently upgrades
the single installed copy, verifies the installed version matches, prunes old
installers and relaunches. Never run release/win-unpacked directly: that is a
build artifact, and running it is how you end up looking at a stale build while
believing it is current.
npm run package produces release/MuchBetter Client Wall Setup <version>.exe
(about 107 MB, installs per-user with no admin prompt). Prefer npm run deploy,
which also installs it.
The wall has no taskbar button and never steals focus, so the tray icon is the control surface. Right-click it for layer mode, click-through, start-with-Windows, and the diagnostics overlay.
| Hotkey | Action |
|---|---|
Ctrl+Alt+W | Focus mode: brings the wall above other windows and enables interaction |
Ctrl+Alt+Shift+W | Show / hide the wall |
Esc | Leave focus mode |
Tab | Full build: switch between Claude Stats and Work (when the wall has the pointer) |
%APPDATA%\MuchBetter Client Wall\settings.json. Every key has a working default,
so deleting the file is safe.
| Key | Default | Notes |
|---|---|---|
layerMode | "desktop" | desktop (below apps, interactive), top (always on top). For a wall that ignores the mouse, set clickThrough |
performance | "auto" | auto inspects the machine. Force with low/balanced/high |
clickThrough | false | The wall ignores the mouse entirely |
opacity | 1 | Wall opacity in ambient mode |
monitors | [] | Empty means all displays |
mode | "stats" | stats or, in the full build, work. The stats build always loads stats; the old clients loads as stats |
workFilter | "mine" | Full build: Work mode's Mine (Owner includes Gustav) or everyone |
mailVip | ["[email protected]"] | Full build: senders whose new mail notifies in quiet hours too. Plain addresses; [] turns VIPs off. The stats build has none |
debug | false | Diagnostics overlay |
askModel | "gemini-3.8-flash-medium" | Full build: the Gemini model Ask Gemini uses, chosen in the tray |
Everything on that view is read from files on this machine: Claude Code's
transcripts (~/.claude/projects/**/*.jsonl, ~/.claude/history.jsonl, plus the
projects folder of any second account in ~/.claude-account<N>, counted into the
same Claude totals), T3
Code's local database as a fallback for Claude sessions missing from them,
Codex's session rollouts (~/.codex/sessions, see
docs/codex-usage.md), and the one plan-limits reading
Claude Code keeps under cachedUsageUtilization in ~/.claude.json (nothing
else in that file, which also holds account and login details, is read into the
app). No API, no account, and no network to produce any of it. The app reaches the network for exactly two things, neither of
which reads or sends a transcript: projection, which is off until
you turn it on, and the update check in the stats build. In the full build,
Ask Gemini also sends a summary of these numbers to Google, but
only when you ask a question, and through the Antigravity CLI rather than the
wall itself.
Panels are tagged MEASURED (straight from the files) or DERIVED (computed, with the method printed on the panel). Cost is derived at API list prices and is not a bill. Plan limits are the last reading Claude Code and Codex saved, shown with its age and never fetched live; a window past its reset time shows as reset rather than its old percent. When a reading is missing the panel is hidden rather than showing a 0.
The first index takes about five seconds and is cached; after that it only reads the bytes appended since last time, triggered by a filesystem watch rather than a timer. Full reasoning in ARCHITECTURE.md.
Full build only. Click ASK top right, press Ctrl+Alt+G, or use the tray, type a question and press Enter: Gemini answers in a panel over the wall, with the wall's own numbers in front of it. Escape cancels or closes. The last five questions stay for the session, in memory only.
This is the beginning of the README. Read the rest at the original. Read the full README ↗