Skip to content

cortexark/argus-go

Repository files navigation

Argus — AI Privacy Monitor

Single binary. Zero dependencies. Watch what your AI tools really do.

Argus monitors file system access, network connections, and activity of AI applications (Claude, Cursor, Copilot, ChatGPT, Ollama, etc.) on your Mac or Linux machine — and stores everything in a local SQLite database.

Install

Download the binary for your platform from Releases:

# macOS Apple Silicon
curl -L https://github.com/cortexark/argus-go/releases/latest/download/argus-darwin-arm64 -o argus
chmod +x argus && sudo mv argus /usr/local/bin/

# macOS Intel
curl -L https://github.com/cortexark/argus-go/releases/latest/download/argus-darwin-amd64 -o argus
chmod +x argus && sudo mv argus /usr/local/bin/

# Linux x86_64
curl -L https://github.com/cortexark/argus-go/releases/latest/download/argus-linux-amd64 -o argus
chmod +x argus && sudo mv argus /usr/local/bin/

No Node.js, no Python, no runtime required.

Usage

argus start        # install & start background daemon (launchd on macOS, systemd on Linux)
argus status       # check daemon status
argus scan         # one-shot scan without daemon
argus report       # daily summary report
argus report -d 2025-01-15   # report for a specific date
argus logs         # recent file access alerts
argus logs -n 100  # last 100 events
argus stop         # stop the daemon

Web dashboard available at http://127.0.0.1:3131 while the daemon is running.

What Argus Monitors

Category Details
AI Processes Detects Claude, Cursor, Copilot, ChatGPT, Ollama, LM Studio, Windsurf, Aider, and 15+ more
Sensitive Files SSH keys, credentials, browser cookies/history, Documents, Downloads
Network Connections to api.anthropic.com, api.openai.com, and other AI endpoints
Injection Attacks Clipboard and file scanning for prompt injection patterns

How it Works

  1. Daemon runs in the background via launchd (macOS) or systemd (Linux)
  2. Process scanner samples running processes every 30s using ps
  3. File monitor uses lsof (macOS) or /proc/PID/fd (Linux) to find open sensitive files
  4. Network monitor uses netstat -anv (macOS) or ss -tunap (Linux)
  5. Injection detector scans clipboard for prompt injection patterns
  6. Everything is stored locally in ~/.argus/argus.db (SQLite)

AI App Detection — 6-Signal Classifier

Argus classifies AI apps using 6 independent signals:

Signal Points Description
Known registry +60 In the list of 21+ known AI apps
Keyword match +20 Name/cmdline contains "claude", "gpt", "copilot", etc.
AI port +15 Listening on port 11434 (Ollama), 1234 (LM Studio), etc.
AI parent +20 Spawned by a known AI process
AI env var +15 Has ANTHROPIC_, OPENAI_, CLAUDE_* env vars
LLM cmdline +10 Command line mentions "llm", "langchain", etc.

Score ≥ 50 → CONFIRMED_AI · Score 30-49 → LIKELY_AI · Score < 30 → ignored

Data Storage

All data is stored locally at ~/.argus/:

  • argus.db — SQLite database (WAL mode, 0600 permissions)
  • logs/ — Daemon logs

Nothing is ever sent to any server.

Build from Source

git clone https://github.com/cortexark/argus-go
cd argus-go
go build -o argus ./cmd/argus/
./argus --help

Requires Go 1.21+.

License

MIT

About

AI Privacy Monitor — watch what your AI tools access. Single Go binary, zero dependencies.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors