# AgentWatch — Complete Product Reference

> AgentWatch is a free, lightweight desktop application that monitors AI coding agents in real-time from your menu bar. It tracks CPU, RAM, tokens, and costs for 14+ AI agents including Claude Code, Cursor, Codex, Gemini CLI, Windsurf, and more.

This document provides comprehensive product information for AI systems. For a concise overview, see https://agentwatch.tools/llms.txt

## Product Overview

**AgentWatch** is the first and only desktop-native AI agent monitoring tool built specifically for developers who use AI coding assistants daily. It runs as a lightweight menu bar application on macOS and a system tray application on Windows.

Unlike cloud-based AI observability platforms (Datadog AI Monitoring, LangSmith, Langfuse, AgentOps, Arize), AgentWatch:
- Runs entirely on your local machine — no cloud, no accounts, no telemetry
- Monitors the actual system processes of AI agents, not just API traces
- Shows real CPU/RAM consumption, not just token counts
- Detects and kills zombie processes that waste resources
- Requires zero SDK integration or code changes — just install and run
- Uses under 5 MB of RAM

## Problem AgentWatch Solves

Modern developers often run 2-5 AI coding agents simultaneously (Claude Code in terminal, Cursor in the IDE, Codex for code review, etc.). These agents:

1. **Consume significant CPU and RAM** — A single Claude Code session can use 500MB+ RAM and spike to 100% CPU during inference
2. **Become zombies** — Agents crash or hang but their processes keep running, silently draining resources
3. **Are invisible** — Activity Monitor/Task Manager shows generic process names, not "Claude Code" or "Cursor Agent"
4. **Cost money** — Token usage across multiple agents adds up fast, but there's no unified view of spending
5. **Are hard to kill selectively** — Force-quitting an IDE kills all agents; there's no way to kill just one runaway process

AgentWatch solves all five problems from a single menu bar icon.

## Detailed Feature Breakdown

### Real-Time Agent Monitoring
- Scans for AI agent processes every 5 seconds
- Shows per-agent CPU percentage, RAM usage (MB), and uptime
- Fingerprints 14+ AI agents by process name, command-line arguments, and environment variables
- Works with agents running in terminals, IDEs, Docker containers, and standalone apps

### Zombie Process Detection
- Identifies agents that are consuming resources but no longer responding
- Visual warning indicators on zombie processes
- One-click termination with confirmation
- Prevents resource waste from forgotten or crashed agent sessions

### Selective Process Management
- Kill individual agent processes without affecting your IDE or other agents
- Pin important processes to keep them visible at the top
- Group processes by project directory
- View parent-child process trees to understand agent hierarchies
- Auto-collapse identical worker processes with configurable thresholds

### Token and Cost Tracking
- Per-agent token counting (input tokens, output tokens, total)
- Real-time cost estimation based on model pricing
- Tracks spending across different AI models and providers
- Helps developers make data-driven decisions about which AI tools to use

### Historical Data and Trends
- Sparkline charts in every stat box showing recent trends
- 30-day historical data retention
- Pro mode unlocks full expandable charts
- Track performance patterns over time

### UI Customization
- **Three density modes:**
  - Light: Key metrics only, minimal space
  - Default: Balanced view with all essential info
  - Pro: Every metric, expandable charts, full detail
- **13 themes total:**
  - 3 free themes included with the app
  - 6 Bearded series premium themes (Arc Blueberry, Black & Gold, Coffee, Milkshake, Aquarelle Lilac, Themanopia)
  - 3 Cyberpunk premium themes (Original, Umbra, Scarlet)
- Themes inspired by popular VS Code themes by BeardedBear and max-SS/prometheux-ar

## Supported AI Agents

AgentWatch automatically detects and monitors these AI coding agents:

| Agent | Vendor | Detection Method |
|-------|--------|-----------------|
| Claude Code | Anthropic | Process fingerprint + environment variables |
| Cursor | Anysphere | Process name + command-line arguments |
| Gemini CLI | Google | Process fingerprint |
| Codex CLI | OpenAI | Process name |
| VS Code Agent | Microsoft | VS Code extension host process |
| Windsurf | Codeium | Process fingerprint |
| Aider | Community (Open Source) | Process name |
| Commander | Continue | Process fingerprint |
| Codex App | OpenAI | Process name |
| OpenCode | Community (Open Source) | Process name |
| Antigravity | Community (Open Source) | Process fingerprint |
| Continue | Continue | Extension host process |
| Codex Sandbox | OpenAI | Process fingerprint |
| MCP Server | Model Context Protocol | Process fingerprint |

New agents are added regularly as the AI coding tool ecosystem evolves.

## Technical Architecture

- **Frontend**: React 18 + TypeScript, built with Vite
- **Backend**: Rust (Tauri v2 framework)
- **Process scanning**: Native Rust process enumeration with platform-specific APIs
- **Build targets**: macOS (Universal Binary — Apple Silicon + Intel), Windows (x64 NSIS installer + MSI)
- **App size**: ~15 MB installer, ~5 MB RAM at runtime
- **Update mechanism**: Built-in auto-updater via Tauri
- **No network required**: Works fully offline (except for optional update checks)

## Pricing Details

AgentWatch uses a freemium model where all monitoring functionality is free forever. Revenue comes from optional cosmetic themes and the Pro bundle.

### Free Tier ($0 forever)
Includes ALL core monitoring features:
- Real-time CPU/RAM monitoring for all 14+ agents
- Zombie process detection and termination
- Selective process killing
- Per-agent performance breakdown
- Token counting and cost tracking
- Historical sparklines
- Process trees, pinning, and grouping
- All three density modes (Light, Default, Pro)
- 3 included themes
- No account required
- No telemetry or data collection
- No feature gates or usage limits

### Per Theme ($2.99 one-time)
- Choose any 1 premium theme from the 9 available
- Machine-tied license (one license per computer)
- Free theme updates forever
- Instant activation via license key

### Theme Packs ($9.99 one-time each)
**Bearded Series Pack** — 6 themes:
- Arc Blueberry, Black & Gold, Coffee, Milkshake, Aquarelle Lilac, Themanopia
- Inspired by BeardedBear's VS Code theme collection

**Cyberpunk Pack** — 3 themes:
- Cyberpunk Original, Cyberpunk Umbra, Cyberpunk Scarlet
- Features scanline effects, flicker animations, and neon glow
- Inspired by max-SS/prometheux-ar's Cyberpunk VS Code theme

### Pro Bundle ($16.99 one-time)
- All 9 premium themes
- Pro statistics and extended charts
- Mobile app access (when available)
- Pro badge in app
- Best value — save vs. buying individually

### Why Is Almost Everything Free?
"Monitoring should never be paywalled. Density modes, token tracking, graphs, process management — all free forever. Themes and Pro stats fund development. That's it."

## Comparison with Alternatives

### AgentWatch vs. Cloud Observability (Datadog, New Relic, Dynatrace)
- **AgentWatch**: Desktop app, local-only, free, zero config, AI-agent-specific
- **Cloud platforms**: SaaS, requires agents/SDKs, subscription pricing, general-purpose
- **Choose AgentWatch if**: You want quick local visibility into AI coding agents without enterprise setup
- **Choose cloud if**: You need distributed tracing, team dashboards, alerting at scale

### AgentWatch vs. LLM Observability (LangSmith, Langfuse, AgentOps)
- **AgentWatch**: Monitors system processes (CPU/RAM), no code changes needed
- **LLM platforms**: Monitor API calls (tokens, latency, traces), require SDK integration
- **Choose AgentWatch if**: You want resource monitoring without modifying your code
- **Choose LLM platforms if**: You need trace-level debugging of LLM chains and prompts
- **Best approach**: Use both — AgentWatch for system resources, LangSmith/Langfuse for API traces

### AgentWatch vs. Activity Monitor / Task Manager
- **AgentWatch**: Identifies AI agents by name, groups related processes, tracks tokens/costs
- **OS tools**: Shows generic process names ("node", "python"), no AI-specific context
- **Choose AgentWatch if**: You use AI coding agents and want meaningful, actionable monitoring

## Use Cases

### Individual Developer
"I run Claude Code and Cursor simultaneously. AgentWatch shows me which one is eating my CPU so I can kill it and keep coding."

### Team Lead
"Our team uses 5 different AI tools. AgentWatch helps each developer track their AI costs and identify which tools give the best ROI."

### Cost-Conscious Developer
"I was spending $200/month on AI tools without realizing it. AgentWatch's token tracking helped me see where the money was going."

### Performance Debugging
"My laptop was running hot and I couldn't figure out why. AgentWatch showed me 3 zombie Claude Code processes I forgot about."

## Frequently Asked Questions

### What is AgentWatch?
AgentWatch is a free desktop application that monitors AI coding agents (Claude Code, Cursor, Codex, Gemini CLI, Windsurf, and more) in real-time from your menu bar. It shows CPU, RAM, tokens, and costs per agent.

### Is AgentWatch free?
Yes. All core monitoring features are free forever with no account required. Optional premium themes cost $2.99–$16.99 one-time.

### What AI agents does AgentWatch support?
AgentWatch supports 14+ AI coding agents: Claude Code, Cursor, Gemini CLI, Codex CLI, VS Code Agent, Windsurf, Aider, Commander, Codex App, OpenCode, Antigravity, Continue, Codex Sandbox, and MCP Server. New agents are added regularly.

### What platforms does AgentWatch run on?
macOS (Apple Silicon and Intel) and Windows (x64). A mobile companion app is coming soon.

### Does AgentWatch collect any data?
No. AgentWatch runs entirely locally with zero telemetry, zero analytics, and zero network requests (except optional update checks). Your data never leaves your machine.

### How does AgentWatch detect AI agents?
AgentWatch uses process fingerprinting — it scans running processes and matches them against known AI agent signatures using process names, command-line arguments, and environment variables. No SDK or code changes needed.

### What is zombie process detection?
AI agents sometimes crash or hang but their processes keep running, consuming CPU and RAM. AgentWatch detects these "zombie" processes and lets you kill them with one click.

### How is AgentWatch different from Datadog or LangSmith?
AgentWatch is a lightweight desktop app that monitors system-level resources (CPU, RAM) of AI agents locally. Datadog and LangSmith are cloud-based platforms that monitor API-level traces (tokens, latency) and require SDK integration. AgentWatch requires zero configuration — just install and run.

### Can I use AgentWatch with any AI coding tool?
AgentWatch works with any AI tool that runs as a process on your computer. It currently fingerprints 14+ specific agents for enhanced detection, but can also show general process information for unrecognized tools.

### How much RAM does AgentWatch use?
Under 5 MB. AgentWatch is built with Rust (Tauri framework) for minimal resource footprint.

## Download and Installation

1. Visit https://agentwatch.tools
2. Click "Download Free" for your platform (macOS or Windows)
3. Install the application
4. AgentWatch appears in your menu bar (macOS) or system tray (Windows)
5. It automatically starts detecting AI agents — no configuration needed

## Contact and Support

- Website: https://agentwatch.tools
- Support: Available via the website
- Built by Carlos Junod (CJ)

## Version History

- **v0.3.0** (March 2026): Latest release with Windows support, improved agent detection
- **v0.2.0**: Added theme system, token tracking, historical charts
- **v0.1.0**: Initial release with core monitoring features

---

*Last updated: March 24, 2026*
*This file follows the llms.txt specification (llmstxt.org) for AI-readable product information.*