VibeRooms
How much room is left on your AI subscriptions? VibeRooms is a small self-hosted service that tracks the usage
limits of your AI subscriptions and API keys (Claude, ChatGPT, GitHub Copilot, OpenRouter, DeepSeek, OpenAI and
Anthropic APIs…, several accounts each): headroom before the next limit, when it resets, whether you will hit it at
the current pace, and which account to use right now. Anything else that exposes its quota over HTTP can be added
from the dashboard, without code.
Français
- Dashboard: per account, the headroom before the next limit, the 5-hour session, the week, per-model caps,
extra-usage credits, a pace marker and a projection ("limit around 19:10 at this pace"), plus history charts.
- "Use now": per tool (Claude Code, Codex), the account with the most headroom.
- macOS menu bar app: one gauge per account, native menu, notifications at 80 % / 95 % / reset.
- Claude Code status line:
Personal 20% ↻1h22 · wk 4% │ Work 31% left.
- Homepage integration: a group of widgets, kept in sync automatically.
- Self-hosted: one Rust binary (≈7 MB) with SQLite. Run it on your own computer, or on an always-on home server
reached over Tailscale to follow it from every device. No cloud, no telemetry.
- English / French user interface.
[!WARNING]
Unofficial. VibeRooms is not affiliated with Anthropic, OpenAI or Google. Subscription limits are not exposed by
public APIs: VibeRooms signs in with the OAuth clients of the official CLIs (Claude Code, Codex) and reads the
internal endpoints those tools use. They can change or break at any time, and using them may not be covered by the
providers' terms. VibeRooms only reads your profile and usage (it never sends prompts). Use it at your own risk,
and read your providers' terms first.
Supported providers
| Service | What is tracked | How | Status |
|---|
| Claude (Pro, Max, Team, Enterprise) | 5-hour session, week, per-model weekly caps, extra credits | Sign in (OAuth, user:profile scope) | Stable |
| ChatGPT (Plus, Pro, Team…), as used by Codex | 5-hour window, week, per-model limits | Sign in (OAuth) | Experimental |
| GitHub Copilot | Premium requests or AI credits this month | gh auth token | Experimental (internal endpoint) |
| Kimi Code | 5-hour window, week, monthly quota | Kimi Code key | Experimental (internal endpoint) |
| GLM Coding Plan (Z.ai) | 5-hour and weekly token windows, monthly tool calls | Coding Plan key | Experimental (internal endpoint) |
| OpenRouter | Key spending limit, today, this month; account credits | API key / Management key | Documented API |
| DeepSeek | Prepaid balance | API key | Documented API |
| Moonshot (Kimi API) | Pay-as-you-go balance | API key | Documented API |
| xAI API (Grok) | Prepaid credits of a team | Management key + team ID | Documented API (ledger updated at cycle close) |
| Anthropic API | Month-to-date spend, against an optional budget | Admin key | Documented API |
| OpenAI API | Month-to-date spend, against the org spend limit | Admin key | Documented API |
| Anything else | Remaining / used / limit / % / reset read from any JSON endpoint | Custom connector | — |
Not possible today: Google AI Pro/Ultra (Google moved individual plans to the closed-source Antigravity CLI,
whose quota API is locked to its client) and SuperGrok / Mistral Le Chat (no API). Gemini API (AI Studio)
quotas are readable through Cloud Monitoring with a service account: planned.
How it works
your laptop(s) home server (Mac mini, NAS, Pi…)
┌───────────────────────┐ Tailscale ┌──────────────────────────────┐ ┌──────────────────┐
│ browser: dashboard │ ───────────► │ viberooms (Rust, SQLite) │ ─────► │ provider usage │
│ menu bar app (macOS) │ ◄─ JSON ──── │ • one OAuth session/account │ every │ endpoints │
│ Claude Code status │ │ • history, pace, projection │ 3 min │ (read-only) │
└───────────────────────┘ │ • Homepage widgets │ └──────────────────┘
└──────────────────────────────┘
Everything can also run on a single machine: the server and the menu bar app both default to 127.0.0.1:8790. A
separate, always-on server keeps the history complete while your laptop sleeps, and lets every device see the same data.
Each account is connected from the dashboard with its own OAuth session (PKCE), requesting the narrowest scope
available (user:profile for Claude). VibeRooms never reuses the tokens of your CLIs: refresh tokens rotate, and
sharing them would log your Claude Code or Codex out. Tokens are stored on the server in accounts.json (mode
0600) and are never sent to the browser or to the menu bar app.
Install the server
Prebuilt binaries: download the archive for your system from the
Releases page (Linux x86_64 and aarch64, macOS Apple Silicon),
then run tar -xzf viberooms-*.tar.gz && ./viberooms. The macOS builds are not notarized by Apple: if macOS refuses to
run one, clear the download flag with xattr -dr com.apple.quarantine viberooms.
From source (Rust 1.80+):
git clone https://github.com/mifefr/viberooms && cd viberooms
cargo build --release -p viberooms
./target/release/viberooms
Then open http://localhost:8790 and click Add account. To reach it from your other devices, also listen on a
private address: VR_HOST=127.0.0.1,100.64.0.10 (localhost + your Tailscale address).
This is the beginning of the README. Read the rest at the original. Read the full README ↗
