Argus

Argus sees everything your AI apps do.

Monitor exactly what your AI applications access on your machine — files, credentials, browser data, and network connections. Argus uses a 6-signal AI detection engine to identify AI processes with surgical precision, then tracks their behavior with system notifications.

Why Argus?

AI desktop applications like Claude, Cursor, ChatGPT, and GitHub Copilot run on your machine with broad permissions. You have tools like Little Snitch to watch network ports and OverSight to monitor camera/microphone access. But nothing tells you exactly which files your AI app just read, whether they contain prompt injection attacks, or if your SSH keys were accessed.

Argus fills that gap with transparent, real-time monitoring built on open-source tooling you can inspect.

Dashboard

Argus runs as a macOS menubar app with a web dashboard at localhost:3131. Here's what it looks like monitoring a live machine with Claude Desktop, Claude Code, Ollama, and Python running.

Privacy Monitoring — Overview

The Overview tab shows everything at a glance: 5 active AI apps detected, 37 file access alerts (credentials, browser data, system files), and real-time activity feed. The Sensitive Access Review panel lets you mark each file access as "Expected" or "Suspicious" — Argus observes and alerts, it does not block. Claude Desktop was caught accessing Safari bookmarks, Keychain databases, SSH agent sockets, and /etc/hosts — all flagged with severity badges (Credentials, Browser Data, Documents, System).

AI Tool Usage Tracking

The Usage tab (inspired by TermTracker) reads local session data from your AI coding tools and calculates what your usage would cost at standard API rates. In this example: 634.6M tokens consumed across 21 Codex sessions, equivalent to $4.4K at API pricing. The per-tool cards show model breakdowns (gpt-5.3-codex: 494M tokens, unknown: 140.7M, gpt-5.4-mini: 12.1K) with color-coded distribution bars. The session table ranks every conversation by token usage so you can see which tasks consumed the most resources. Supports Codex (SQLite), Claude Code (JSONL metrics), and Cursor (SQLite tracking DB).

File Access Alerts

The File Alerts tab shows every sensitive file access by AI apps with timestamps, file paths, and severity classification. Each row shows exactly when (11:05:10 PM), which app (Claude Desktop), what file (/Users/.../keychain-2.db), and what category (Credentials, Browser Data, Documents, System). This is the raw evidence trail — every time an AI app touches your SSH keys, Keychain, Safari data, or system files, it's logged here.

The Problem

Claude might read /Documents/secret-project.md — does it contain sensitive data?
Cursor connected to api.openai.com — or is it api-exfil.com (typosquatting)?
ChatGPT accessed ~/.ssh/id_rsa — credential theft risk.
A Node.js process claims to be an AI agent — how can you verify?

Argus detects and alerts on all of these with specific, actionable notifications.

What Argus Detects

Signal	Detection Method	Risk Level
Process Ancestry	Walks parent process tree for known AI apps (Claude, Cursor, ChatGPT, etc.)	Definitive
MCP Server Pattern	Detects stdin/stdout pipes → MCP server spawned by AI orchestrator	High confidence
AI Keywords	Scans command line for `claude`, `langchain`, `openai`, `copilot`, `agent`, etc.	Medium confidence
Network Endpoints	Monitors connections to Anthropic, OpenAI, Google, Mistral, Ollama APIs	Definitive
TCC Permissions	Checks if process has Full Disk Access on macOS	Escalation indicator
Code Signing	Verifies code-signing authority on binaries (Anthropic, OpenAI, GitHub, etc.)	Definitive

Score calculation:

Score ≥ 50 → CONFIRMED_AI (at least one definitive signal)
Score 30–49 → LIKELY_AI (multiple confidence signals)
Score < 30 → NOT_AI (background noise filtered out)

File Access Monitoring

When an AI app reads sensitive files, Argus classifies by risk level:

🔑 Credentials — .ssh/, .aws/, .gnupg/, 1Password, Bitwarden
🌐 Browser Data — Chrome/Firefox/Safari passwords, cookies, history
📄 Documents — Files in ~/Documents, ~/Downloads, ~/Desktop
⚙️ System Files — /etc/passwd, .env, .npmrc

Browser Automation Detection

Argus detects four types of browser access:

Direct CDP connections — AI agents connecting to Chrome on port 9222
Browser process spawning — AI apps launching (headless) Chrome/Firefox
AppleScript control — Using osascript to automate Safari or Chrome
Browser extension calls — Browser extensions making requests to AI APIs

Notifications

Every alert tells you exactly what happened:

Argus — Claude
🔑 Credentials access detected
Accessed credential file: ~/.ssh/id_rsa
SSH key / cloud credential risk.

---

Argus — Cursor
new_connection
Connected to api.openai.com  port 443

---

Argus — node (AI agent)
⚙️ System Files
Read 3 sensitive files
1× 🔑 Credentials, 2× 📄 Documents

Throttling: Max 1 notification per (app + alert type) per 5 minutes. Batching: If 3+ file alerts fire in quick succession, they're grouped into one summary.

Real-World Example

Here's Argus running on a live machine during an active Claude Code session:

$ argus status

Argus Status
=================
  Service:  RUNNING
  PID:      69546
  Daemon:   RESPONDING
  Uptime:   6s
  Memory:   94MB

Last 24h Summary
----------------
  Processes seen:   3
  File alerts:      21
  Network events:   5,265

$ argus heatmap

Access Heatmap (last 24h)
────────────────────────────────────────────────────────────────────────────────
~/Library/Keychains                       ██████████████████████████████  (14)
~/Library/Keychains/6DB3F6AA-C887-.../   ███████████████                 (7)
────────────────────────────────────────────────────────────────────────────────

What was found:

File	Times	Severity	Who
`~/Library/Keychains/login.keychain-db`	14×	CRITICAL	Claude Code
`~/Library/Keychains/6DB3F6AA-.../keychain-2.db-wal`	7×	CRITICAL	Claude Code

Network connections detected:

App	Remote	Port	Service
Claude (Anthropic)	`2607:6bc0::10`	443	Anthropic CDN
Claude (Anthropic)	`34.149.66.137`	443	Google Cloud
Claude (Anthropic)	`160.79.104.10`	443	Cloudflare
Claude (Anthropic)	`172.65.251.78`	443	Cloudflare
Claude (Anthropic)	`2606:50c0:8003::154`	443	GitHub CDN

$ argus export --format csv

timestamp,app,event_type,detail,severity
2026-03-18T01:22:19Z,Claude (Anthropic),FILE,~/Library/Keychains/login.keychain-db,CRITICAL
2026-03-18T01:10:04Z,Claude (Anthropic),FILE,~/Library/Keychains/login.keychain-db,CRITICAL
2026-03-18T00:54:23Z,Claude (Anthropic),FILE,~/Library/Keychains/keychain-2.db-wal,CRITICAL
2026-03-18T00:43:16Z,Claude (Anthropic),NET,api.anthropic.com,INFO
...

The Keychain reads are Claude Code fetching git/SSH credentials. Network traffic is all Anthropic/Cloudflare/GitHub — nothing suspicious. This is exactly the kind of transparency Argus provides.

Quick Install

macOS & Linux

npm install -g argus-monitor
argus install
argus start
argus status

One-liner

curl -fsSL https://raw.githubusercontent.com/cortexark/argus/main/install.sh | bash

The installer will:

Verify Node.js ≥ 18 is installed
Install Argus globally via npm
Register it as a system service (LaunchAgent on macOS, systemd on Linux)
Start the daemon
Print command reference

First Launch Permission Message (macOS app)

On first launch, Argus now shows a clear choice:

Start in Basic Mode (Recommended) — process + network monitoring, reduced permissions
Enable Deep Monitoring — includes cross-app file access detection (may trigger the macOS prompt:
"Argus" would like to access data from other apps.)

Argus runs locally on your Mac by default.

Commands

Service Management

argus install              # Register Argus as a persistent system service
argus uninstall            # Remove system service
argus start                # Start monitoring daemon
argus stop                 # Stop monitoring daemon
argus restart              # Restart monitoring daemon
argus status               # Show daemon status + event counts

Live Monitoring

argus watch                # Launch interactive TUI dashboard
argus logs -f              # Follow live logs (tail -f style)

Log Viewing

argus logs [options]
  --follow / -f            Follow output (tail -f style)
  --lines N / -n N         Show last N lines (default 50)
  --since <duration>       e.g., "1h", "30m", "2d"
  --level <level>          Filter by level: trace|debug|info|warn|error
  --json                   Raw JSON output

# Examples
argus logs -f
argus logs --since 1h --level warn
argus logs --json | jq '.msg'

Reporting

argus report [options]
  --since <ISO>            Filter events since timestamp (ISO 8601)
  --process <name>         Filter to specific process
  --alerts-only            Show only file access alerts
  --format json            Output JSON instead of text

# Examples
argus report --since 2024-01-01T00:00:00Z
argus report --process Claude --format json
argus report --alerts-only

macOS Permissions Check

argus tcc                  # Check which AI apps have Full Disk Access
                           # (macOS only — shows TCC database entries)

How It Works

Architecture Diagram

┌─────────────────────────────────────────────────────────────┐
│ Argus Daemon (runs in background)                           │
├─────────────────────────────────────────────────────────────┤
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Process Scanner (every 5s)                           │  │
│  │ ├─ ps -A → process tree                             │  │
│  │ └─ 6-signal AI classifier per process               │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ File Monitor (chokidar)                              │  │
│  │ └─ Watch AI processes → file access events          │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Network Monitor (lsof)                               │  │
│  │ └─ Enumerate AI process connections                 │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Browser Monitor                                      │  │
│  │ ├─ Detect CDP (Chrome DevTools) connections         │  │
│  │ ├─ Detect browser spawning                          │  │
│  │ ├─ Detect AppleScript browser control               │  │
│  │ └─ Detect browser extension API calls               │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ SQLite Event Store (~/.argus/data.db)               │  │
│  │ ├─ process_classifications                          │  │
│  │ ├─ file_events                                      │  │
│  │ ├─ network_events                                   │  │
│  │ └─ browser_events                                   │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
│  ┌──────────────────────────────────────────────────────┐  │
│  │ Notifier (throttled + batched)                       │  │
│  │ └─ Native OS notifications → user                    │  │
│  └──────────────────────────────────────────────────────┘  │
│                                                              │
└─────────────────────────────────────────────────────────────┘

                          ↓ IPC socket

                 ┌─────────────────────┐
                 │ Argus CLI           │
                 ├─────────────────────┤
                 │ status, logs,       │
                 │ watch, report, tcc  │
                 └─────────────────────┘

Data Flow

Discovery — Process scanner runs every 5 seconds, building a process tree via ps -A
Classification — Each AI process is scored on 6 signals (ancestry, pipes, keywords, network, TCC, code signing)
File Monitoring — Chokidar watches known AI app working directories for file access
Network Monitoring — lsof enumerates active connections from AI processes
Storage — All events written to SQLite at ~/.argus/data.db (never leaves your machine)
Notification — Throttled, batched alerts sent as native OS notifications
Reporting — CLI generates human-readable or JSON reports from the event database

Supported AI Apps

App	Category	Process Name
Claude Code	AI Code Editor	`claude`
Claude Desktop	LLM Desktop	`Claude`
OpenAI Codex	AI Code Editor	`codex`
ChatGPT	LLM Desktop	`ChatGPT`, `chatgpt`
Cursor	AI Code Editor	`cursor`, `Cursor`
VS Code + Copilot	AI Code Editor	`Code`, `code`
Windsurf	AI Code Editor	`windsurf`, `Windsurf`
Continue.dev	AI Code Editor	`continue`
Perplexity	LLM Desktop	`Perplexity`
Ollama	Local LLM	`ollama`
LM Studio	Local LLM	`LM Studio`, `lmstudio`
GitHub Copilot	AI Assistant	`copilot`
Amazon Q	AI Assistant	`amazonq`, `Amazon Q`
Tabnine	AI Code Editor	`tabnine`
Node.js agents	Runtime	`node`
Python agents	Runtime	`python`, `python3`

Detected AI endpoints:

Anthropic Claude API
OpenAI API (ChatGPT, GPT-4, etc.)
Google Gemini API
Mistral AI
Cohere
Together AI
OpenRouter
Hugging Face
Replicate
Codeium/Windsurf
GitHub Copilot
Perplexity AI

Data Storage

All monitoring data is stored in an SQLite database at ~/.argus/data.db. This database is:

Local-only — Never leaves your machine or is sent to any server
Permanent — Survives daemon restarts, reboots, and app updates
Queryable — Can be inspected directly with sqlite3
Portable — Copy ~/.argus/ to backup or analyze on another machine

Logs are written to ~/.argus/logs/ (rotating, max 10 MB per file).

Requirements

Node.js ≥ 18.0.0
macOS 12+ (Monterey or later) or Linux (Ubuntu 20+, Debian 11+, etc.)
No root required — Runs as a user LaunchAgent (macOS) or systemd service (Linux)
SQLite — Built-in on all supported platforms
lsof — Pre-installed on macOS; install on Linux: sudo apt-get install lsof

Development

Install Dependencies

npm install

Run Tests

npm test

All tests are in /tests. The test suite includes:

Unit tests for classifiers, monitors, and utilities
Integration tests for database operations
E2E tests for CLI commands
640+ tests, 0 failures

Test Coverage

# View test output with details
node tests/run.js

Tests are run on Node 18, 20, and 22 in CI.

Project Structure

argus/
├── src/
│   ├── index.js                    # Main daemon entry point
│   ├── cli.js                      # CLI command dispatcher
│   ├── ai-apps.js                  # Registry of AI apps + detection config
│   ├── lib/
│   │   ├── config.js               # Frozen config object
│   │   ├── platform.js             # macOS/Linux detection
│   │   ├── exec.js                 # Async command executor
│   │   └── logger.js               # Pino logger setup
│   ├── monitors/
│   │   ├── process-classifier.js   # 6-signal AI detection engine
│   │   ├── process-scanner.js      # Process tree builder
│   │   ├── file-monitor.js         # Watch file access
│   │   ├── network-monitor.js      # Track network connections
│   │   ├── browser-monitor.js      # Detect browser automation
│   │   └── usage-tracker.js        # AI tool usage (Codex/Claude/Cursor)
│   ├── db/
│   │   ├── schema.js               # SQLite schema init
│   │   └── store.js                # Database CRUD operations
│   ├── notifications/
│   │   └── notifier.js             # Throttled + batched alerts
│   ├── daemon/
│   │   ├── daemon-manager.js       # Main event loop
│   │   ├── launchd.js              # macOS LaunchAgent setup
│   │   └── ipc-client.js           # IPC for CLI commands
│   ├── cli/
│   │   └── commands/               # Command implementations
│   ├── dashboard/
│   │   └── dashboard.js            # TUI dashboard (blessed)
│   └── report/
│       └── report-generator.js     # Report formatting
├── tests/
│   ├── run.js                      # Test runner
│   └── [test files]
├── install.sh                      # One-liner installer
└── package.json

Adding a New AI App

Add entry to AI_APPS in src/ai-apps.js:

'myapp': { name: 'My AI App', category: 'AI Code Editor' },

Add any app-specific keywords to CMD_KEYWORDS in src/monitors/process-classifier.js
Add any app-specific API endpoints to AI_ENDPOINTS in src/ai-apps.js
Add tests to ensure detection works

Adding a New Injection Pattern

Edit SENSITIVE_PATHS in src/ai-apps.js to add file paths or credential locations to monitor.

Adding a New Notification Type

Edit src/notifications/notifier.js and add a function to the notify object:

export const notify = {
  myAlert(appName, detail) {
    return sendAlert(appName, 'my_alert', `Custom message: ${detail}`);
  },
};

Security & Privacy

How Argus Stays Safe

No internet access — Argus is purely local. Data never leaves your machine.
Open source — All monitoring logic is visible in this repository.
No telemetry — No analytics, crash reporting, or tracking.
SQLite-only — Events stored in a local database file you can inspect.
Read-only — Argus only reads process information; it doesn't modify anything.

How to Verify Argus Isn't Malicious

Inspect src/ — all networking code is in monitors/network-monitor.js
Check src/notifications/notifier.js — notifications only use node-notifier (native OS API)
Search for http or https — none in production code
Run npm ls to inspect all dependencies (all are production-ready open-source packages)

Uninstall

argus uninstall
npm uninstall -g argus-monitor
rm -rf ~/.argus

Troubleshooting

Daemon won't start

argus logs --level error

Check if Node.js is installed and in PATH:

which node
node --version

No notifications appearing

Check if daemon is running: argus status
Check macOS notifications settings — Argus needs permission to notify
Test manually: argus logs -f to see events in real-time

Too many notifications

Adjust throttle interval in src/lib/config.js:

NOTIFICATION_THROTTLE_MS: 300000,  // 5 minutes

High CPU usage

Reduce scan intervals in src/lib/config.js:

SCAN_INTERVAL_MS: 5000,              // process scan every 5s
FILE_MONITOR_INTERVAL_MS: 3000,      // file monitor every 3s
NETWORK_MONITOR_INTERVAL_MS: 3000,   // network scan every 3s

License

MIT — See LICENSE for details.

Contributing

See CONTRIBUTING.md for guidelines on:

Adding new AI apps
Adding new injection patterns
Creating new notification types
Running and writing tests

Claude Desktop MCP Extension

Argus ships with a built-in MCP server that integrates directly with Claude Desktop. Ask Claude questions about what AI agents are doing on your machine — in natural language.

Setup

Add to your Claude Desktop config (~/Library/Application Support/Claude/claude_desktop_config.json):

{
  "mcpServers": {
    "argus": {
      "command": "node",
      "args": ["/path/to/argus/src/mcp/server.js"]
    }
  }
}

Restart Claude Desktop. Argus tools appear automatically.

Available Tools

Tool	Description
`get_active_ai_processes`	List AI agents detected in the last N hours
`get_file_accesses`	Sensitive file access alerts (credentials, browser data)
`get_network_activity`	Network connections to AI endpoints
`get_sessions`	Session history with start/stop/duration
`get_daily_summary`	Daily stats for any date
`get_injection_alerts`	Prompt injection detections
`get_ai_usage`	Token usage and estimated API costs
`get_monitoring_status`	Overall Argus status and stats

All tools are read-only — they query the local SQLite database and never modify anything.

Example 1: Check what's running

Prompt: "What AI agents ran on my machine today?"

Response:

[
  { "name": "claude", "app_label": "Claude Code (CLI)", "category": "AI Code Editor" },
  { "name": "Claude", "app_label": "Claude Desktop", "category": "LLM Desktop" },
  { "name": "ollama", "app_label": "Ollama", "category": "Local LLM" }
]

Example 2: Check for credential access

Prompt: "Did any AI app access my SSH keys or credentials?"

Response:

[
  {
    "process_name": "Claude",
    "app_label": "Claude Desktop",
    "file_path": "/Users/you/.ssh/id_rsa",
    "sensitivity": "credentials",
    "is_alert": 1,
    "timestamp": "2026-04-03T14:22:19.000Z"
  },
  {
    "process_name": "Claude",
    "app_label": "Claude Desktop",
    "file_path": "/Users/you/Library/Keychains/login.keychain-db",
    "sensitivity": "credentials",
    "is_alert": 1,
    "timestamp": "2026-04-03T14:10:04.000Z"
  }
]

Example 3: Daily summary

Prompt: "Give me a summary of AI activity for yesterday"

Response:

{
  "date": "2026-04-02",
  "processCount": 4,
  "fileAlertCount": 21,
  "networkEventCount": 5265,
  "topPorts": [443, 80, 9222],
  "aiServicesHit": ["Anthropic", "OpenAI", "GitHub Copilot"]
}

Privacy Policy

Argus is a local-only privacy monitoring tool. This policy covers both the Argus application and its MCP extension for Claude Desktop.

Data Collection

Argus monitors AI application activity (processes, file accesses, network connections) on your local machine
All monitoring data is stored in a local SQLite database at ~/.argus/data.db
The MCP extension reads from this same local database

Data Transmission

No data is ever transmitted to any external server, cloud service, or third party
The MCP extension communicates only with Claude Desktop via local STDIO (standard input/output)
There is no telemetry, analytics, crash reporting, or tracking of any kind
The application makes zero outbound network connections

Data Storage

All data remains on your local filesystem at ~/.argus/
Database file permissions are set to owner-only (mode 0600)
You can inspect, export, or delete the database at any time
Old events are automatically cleaned up after 7 days

Data Sharing

Argus does not share data with anyone
The MCP extension provides data only to Claude Desktop running on the same machine
No third-party services, APIs, or SDKs are used for data processing

Your Control

You can pause/resume monitoring at any time via the tray menu
You can delete all data by removing ~/.argus/
All source code is open source and auditable at https://github.com/cortexark/argus
Uninstall instructions: argus uninstall && npm uninstall -g argus-monitor

Contact

For privacy questions or concerns, open an issue at https://github.com/cortexark/argus/issues

Related Tools

Little Snitch — Network firewall (port-level monitoring)
OverSight — Camera/microphone access monitor
Lulu — Open-source firewall
Santa — Google's binary authorization tool

Argus complements these tools by focusing specifically on what AI applications access on your machine.

Stay curious. Stay in control.

Name		Name	Last commit message	Last commit date
Latest commit History 36 Commits
.github		.github
.pm		.pm
.project		.project
docs		docs
electron		electron
homebrew		homebrew
releases		releases
scripts		scripts
src		src
tests		tests
.gitignore		.gitignore
.mcpbignore		.mcpbignore
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
SECURITY.md		SECURITY.md
install.sh		install.sh
manifest.json		manifest.json
package.json		package.json

Folders and files

Latest commit

History

Repository files navigation

Argus

Why Argus?

Dashboard

Privacy Monitoring — Overview

AI Tool Usage Tracking

File Access Alerts

The Problem

What Argus Detects

File Access Monitoring

Browser Automation Detection

Notifications

Real-World Example

Quick Install

macOS & Linux

One-liner

First Launch Permission Message (macOS app)

Commands

Service Management

Live Monitoring

Log Viewing

Reporting

macOS Permissions Check

How It Works

Architecture Diagram

Data Flow

Supported AI Apps

Data Storage

Requirements

Development

Install Dependencies

Run Tests

Test Coverage

Project Structure

Adding a New AI App

Adding a New Injection Pattern

Adding a New Notification Type

Security & Privacy

How Argus Stays Safe

How to Verify Argus Isn't Malicious

Uninstall

Troubleshooting

Daemon won't start

No notifications appearing

Too many notifications

High CPU usage

License

Contributing

Claude Desktop MCP Extension

Setup

Available Tools

Example 1: Check what's running

Example 2: Check for credential access

Example 3: Daily summary

Privacy Policy

Data Collection

Data Transmission

Data Storage

Data Sharing

Your Control

Contact

Related Tools

About

Resources

License

Contributing

Security policy

Uh oh!

Stars

Watchers

Forks

Releases 4

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages