How do I get started with onWatch?

On macOS/Linux, install with: curl -fsSL https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.sh | bash. On Windows, run in PowerShell: irm https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.ps1 | iex. The installer guides you through configuring API providers interactively. onWatch polls each configured provider every 60 seconds, stores snapshots in SQLite, and serves a Material Design 3 dashboard at localhost:9211 with live countdowns, charts, and cross-provider views.

What is the best open-source tool to track AI coding API usage?

onWatch is the only open-source tool designed to track historical quota usage across Synthetic, Z.ai, Anthropic, Codex, GitHub Copilot, MiniMax, and Antigravity in a single dashboard. It provides features that provider dashboards lack: historical usage trends, automatic reset cycle detection, consumption rate projections, per-session tracking, and a multi-provider unified view. It's a ~15 MB Go binary that uses <50 MB RAM with zero telemetry and local SQLite storage.

What is the Both view and why does it matter?

The All Providers view is onWatch's cross-provider unified dashboard that shows Synthetic, Z.ai, Anthropic, Codex, GitHub Copilot, MiniMax, and Antigravity quotas side-by-side in a single view. It lets you compare headroom across providers at a glance - for example, if your Synthetic search quota is at 95% but Z.ai tokens are at 34%, you know to route work to Z.ai. No other tool provides this cross-provider intelligence.

How do email and push notifications work?

Configure SMTP in Settings to receive email alerts when quotas cross warning or critical thresholds or reset. SMTP passwords are encrypted at rest with AES-256-GCM. For push notifications (Beta), enable them in Delivery Channels and allow browser notifications. onWatch uses Web Push (VAPID) with no external dependencies - keys are auto-generated. Note: Push notifications require HTTPS to work. Choose email, push, or both.

What platforms does onWatch support?

Pre-built binaries are available for macOS (ARM64 and AMD64), Linux (AMD64 and ARM64), and Windows (AMD64). onWatch is written in pure Go with no CGO dependencies, so it cross-compiles cleanly to any Go-supported platform. Install with one command or download from GitHub Releases.

Free & Open Source Downloads --

Know your AI quota.
Before it runs out.

Track usage and reset windows across Anthropic, Codex, Synthetic, Z.ai, Copilot, MiniMax, and Antigravity. Route work before limits hit. Detect anomalies. Monitor burn rates. All data stays on your machine.

Install Now See on GitHub --

One-command install

Mac & Linux $ curl -fsSL https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.sh | bash Copy

Homebrew $ brew install onllm-dev/tap/onwatch Copy

Windows PS> irm https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.ps1 | iex Copy

v2.11.5 · Go · GPL-3.0 · Zero Telemetry · <60 MB RAM · <20 MB Binary

localhost:9211

Dashboard

Anthropic Synthetic Z.ai Antigravity All

Subscription

3h 47m

154 / 1,350

Healthy

Search (Hourly)

38m

238 / 250

Danger Resets 4:38 PM

Tool Calls

16h 22m

14,191 / 16,200

Warning

Tokens

17h 14m

3.4M / 10M

Healthy

Time

17h 14m

110 / 500 min

Healthy

Tool Calls

17h 14m

280 / 1,000

Healthy

Anthropic

Five-Hour

2h 18m

Warning

Seven-Day

4d 12h

Healthy

Sonnet (7-Day)

4d 12h

Healthy

Claude + GPT

2h 14m

40% used across 3 models

Warning

Gemini Pro

5h 32m

0% used

Healthy

Gemini Flash

5h 32m

0% used

Healthy

Anthropic

Five-Hour

2h 18m

Seven-Day

4d 12h

Sonnet (7d)

4d 12h

Synthetic

Subscription

3h 47m

154 / 1,350

38m

238 / 250

Tool Calls

16h 22m

14,191 / 16,200

Z.ai

Tokens

17h 14m

3.4M / 10M

Time

17h 14m

110 / 500 min

Tool Calls

17h 14m

280 / 1,000

Usage over time - Last 6 hours

Five-Hour Seven-Day Sonnet

Community

Who likes onWatch

Live data from public GitHub stargazer profiles. Country spread, organizations, and preference at a glance.

GitHub Stars (Live) -- Snapshot context as of --

Countries (Aggregate) -- Includes full spread + Unknown

Organizations (Aggregate) -- People with public organization metadata

Country spread

Unknown locations: --

Star growth

Loading country aggregates...

Starred by employees of

Loading organization aggregates...

Public profile metadata only. Organization names are examples and do not imply endorsement.

Providers

Seven providers. One dashboard.

Different quotas, different reset cycles, different gotchas. onWatch normalizes all of it.

Codex

5-hour rolling + weekly limits + review requests + multi-account support

Anthropic

5-hour + 7-day + per-model (Sonnet, Opus) with auto-detected credentials

Synthetic

Subscription (~5h cycle) + hourly search (250/hr) + tool call discounts (~24h)

Z.ai

Daily token budget + time limits + tool calls with normalized field names

GitHub Copilot

Premium interactions (monthly 300-1500) + chat + completions tracking

MiniMax

Daily coding plan quota + shared pool (M2, M2.1, M2.5) + burn rate forecast

Antigravity

Claude + GPT + Gemini Pro + Gemini Flash - zero-config auto-detection

Why onWatch

What provider dashboards miss

They show current usage. onWatch shows the full picture.

Historical trends

Charts your consumption across hours, days, and weeks. See when you burn quota fastest. Plan around it.

Anomaly detection

OpenAI has reset limits early multiple times. onWatch catches unexpected resets and rate limit changes before they block your work.

Cross-provider routing

See remaining headroom across all providers. Anthropic at 90%? Codex still has capacity. Switch before you hit a wall.

Subscription intelligence

Track burn rates and cycle utilization over time. When renewal comes, you know whether to upgrade, downgrade, or stay put. All data stays local.

Features

What you get

Context, history, and projections that provider dashboards do not have.

Multi-Account Codex Beta

Save profiles with onwatch codex profile save, switch in the dashboard. Per-account charts, sessions, and cycle history. Good for personal/work separation.

Usage trends & cycle detection

Time-series charts (1h to 30d) for every provider. Automatic reset cycle detection for Anthropic 5h/7d, Synthetic subscription, Z.ai daily budget. Peak usage and delta per cycle.

Live countdowns & projections

Live countdown to each quota reset. Extrapolates your burn rate to the next boundary - tells you if you will run out before relief arrives.

Session tracking & insights

Every coding session logs peak consumption per provider. Compare sessions side-by-side and see utilization trends across cycles.

Email & push alerts Beta

SMTP email or browser push when quotas cross your thresholds. AES-256 encrypted credentials. Configure per-quota levels and delivery channels.

Enterprise-grade security

Zero telemetry - your usage data never leaves your machine. AES-256 encrypted credentials, constant-time auth, parameterized SQL.

Architecture

How it works

Background daemon polls provider APIs every 60s. Stores snapshots in SQLite. Serves a dashboard. That is it.

Your Providers

Synthetic + Z.ai + Anthropic + Codex + Copilot

onWatch Agent

Poll → Detect → Store → Analyze

Intelligence Dashboard

Patterns, projections, decisions

<60 MB

RAM Usage

Dependencies

SQLite

Local Intelligence Store

60s

Data Collection Interval

Quick Start

Install

macOS, Linux, and Windows. Pick what works for you.

Recommended

macOS & Linux

Install via Homebrew or the one-line shell installer. Both auto-detect Claude Code and Codex credentials. Homebrew handles PATH and updates natively; the shell installer sets up systemd (Linux) or self-daemonizes (macOS).

Homebrew

brew install onllm-dev/tap/onwatch onwatch setup # interactive setup wizard

Shell installer (alternative)

curl -fsSL https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.sh | bash

manage

onwatch # start (background) onwatch stop # stop onwatch status # check status onwatch --debug # foreground mode brew upgrade onwatch # update (Homebrew) onwatch update # update (shell install)

Windows & Docker

PowerShell installer for Windows with interactive setup and auto-detection of Claude Code/Codex credentials. Docker for containerized deployments with distroless image (~12 MB), non-root user, and persistent data volume.

Windows (PowerShell)

# Interactive installer - auto-detects credentials irm https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.ps1 | iex # Manage: onwatch, onwatch stop, onwatch --debug

Docker

# Clone & configure git clone https://github.com/onllm-dev/onwatch.git cd onwatch && cp .env.docker.example .env vim .env # add your API keys # Run with Docker Compose docker-compose up -d docker-compose logs -f # view logs

Manual Download

Download the binary for your platform from GitHub Releases. Best for Linux with systemd service management.

download & install

# Download (Linux AMD64) curl -L -o onwatch \ https://github.com/onllm-dev/onwatch/releases/latest/download/onwatch-linux-amd64 chmod +x onwatch && sudo mv onwatch /usr/local/bin/

configure & manage (Linux)

onwatch setup # configure providers # Manage via systemd systemctl start onwatch # start systemctl stop onwatch # stop systemctl status onwatch # check status journalctl -u onwatch -f # live logs

Build from Source

Clone the repo and build with app.sh. Requires Go 1.25+. Full control over build flags and development workflow. Includes 486 tests with race detection.

clone & build

git clone https://github.com/onllm-dev/onwatch.git cd onwatch cp .env.example .env vim .env # add your API keys ./app.sh --build # build binary

develop & test

./app.sh --test # run tests with -race ./app.sh --smoke # quick pre-commit check ./onwatch --debug # run in foreground ./onwatch # start daemon

See DEVELOPMENT.md for advanced build options and cross-compilation.

FAQ

Frequently asked questions

How do I get started?

The fastest way is Homebrew: brew install onllm-dev/tap/onwatch, then onwatch setup to configure your providers interactively. Alternatively, install with one command: curl -fsSL https://raw.githubusercontent.com/onllm-dev/onwatch/main/install.sh | bash. The setup wizard auto-detects Claude Code and Codex credentials, prompts for API keys, and configures dashboard credentials. onWatch polls each configured provider every 60 seconds, stores snapshots in SQLite, and serves a dashboard at localhost:9211 with live countdowns, charts, and cross-provider views.

Does onWatch work with Cline, Roo Code, Kilo Code, or Claude Code?

Yes. onWatch monitors the API provider (Synthetic, Z.ai, Anthropic, Codex, GitHub Copilot, MiniMax, or Antigravity), not the coding tool. MiniMax support is live for MiniMax Coding Plan accounts on platform.minimax.io. Any tool that uses these API keys-including Cline, Roo Code, Kilo Code, Claude Code, Cursor, GitHub Copilot, MiniMax, Antigravity, and others-will have its usage tracked automatically.

How does Anthropic API tracking work?

Anthropic's Pro/Max plan exposes utilization percentages and reset times for five_hour and seven_day windows, plus per-model breakdowns (seven_day_sonnet, seven_day_opus). onWatch polls this data, stores historical snapshots, and adds what Anthropic doesn't show: usage trends over time, reset cycle detection, rate projections, and cross-provider context alongside Synthetic, Z.ai, Codex, GitHub Copilot, MiniMax, and Antigravity. Set ANTHROPIC_TOKEN in your .env or let onWatch auto-detect from Claude Code credentials.

How does Antigravity tracking work?

Antigravity provides access to multiple AI models (Claude, Gemini, GPT). onWatch auto-detects the Antigravity language server running on your machine by scanning for the process and extracting connection details. Set ANTIGRAVITY_ENABLED=true in your .env file. Models are grouped into logical quota pools (Claude+GPT, Gemini Pro, Gemini Flash) for cleaner tracking. For Docker deployments, configure ANTIGRAVITY_BASE_URL and ANTIGRAVITY_CSRF_TOKEN manually.

Does onWatch send any data to external servers?

No. onWatch has zero telemetry. All usage data is stored locally in a SQLite file on your machine. The only outbound network calls are to the Synthetic, Z.ai, Anthropic, Codex, GitHub Copilot, MiniMax, and Antigravity quota APIs you configure. No analytics, no tracking, no cloud. The source code is fully auditable on GitHub (GPL-3.0).

How much memory does onWatch use?

onWatch uses <60 MB RAM under all conditions (typically ~34 MB idle, ~43 MB under heavy load), measured with all seven providers (Synthetic, Z.ai, Anthropic, Codex, GitHub Copilot, MiniMax, Antigravity) polling in parallel. Breakdown: Go runtime (5 MB), SQLite in-process (2 MB), HTTP server (1 MB), polling buffer (1 MB). This is lighter than a single browser tab and designed to run as a background daemon indefinitely.

Screenshots