CLGMNI：将 Gemini 大模型能力注入终端的 Agentic AI 工具

CLGMNI is a terminal AI assistant built on Google Gemini, designed for network security experts, system administrators, and advanced users. It functions as an autonomous Agent (with user approval) to execute system commands, supporting features like file context injection, Google search enhancement, session logging, and cost-controlled API management. This tool integrates Gemini's capabilities deeply into terminal workflows, balancing power and security.

CLGMNI (Command Line Gemini) was created by Jay Hawkins to address the need for a terminal-native AI tool. Its "terminal-first" design aligns with the workflows of terminal users (e.g., pipe operations, log analysis). Supported environments include Linux/macOS and Windows via WSL. Unlike web-based AI tools, it allows direct pipe of command outputs and attachment of local files as context for more informed responses.

The standout feature of CLGMNI is its Agent mode (activated via -a). In this mode, Gemini acts as an interactive operator: it analyzes tasks, proposes Bash commands, waits for user approval, executes commands, and analyzes results. For example, it can check network interfaces and listening ports. Security is prioritized: commands are not auto-executed (highlighted in red for user confirmation), and sudo commands use local sudo -v for authentication (no root password is passed to AI).

CLGMNI supports dual API keys (paid GEMINI_API_KEY_PAID and free GEMINI_API_KEY_FREE), with smart fallback to free if paid is unavailable. Users can force free mode with --free to control costs. Context injection features include pipe input of stdout and -f for multiple files (skipping >5MB files to stay within token limits). Google search enhancement (-g) provides real-time data to reduce hallucinations, and -s outputs source URLs.

CLGMNI includes a token bucket rate limiter (300 RPM) and exponential backoff to avoid 429/503 errors. It supports multiple Gemini models: g25f (default 2.5 Flash), g25f-l (2.5 Flash-Lite), g25p (2.5 Pro), and g31p (3.1 Pro-Preview). Users can choose models based on task complexity to balance performance and cost.

CLGMNI is useful in several scenarios:

1. Log Analysis: Pipe logs (e.g., SSH brute force attempts) for summary: cat /var/log/auth.log | grep "Failed password" | python3 CLGMNI.py -m g25f-l "Summarize these SSH brute force attempts."
2. Config Diagnosis: Attach config files and logs (e.g., Nginx 502 errors) for analysis.
3. Threat Intelligence: Use Google search to get latest TTPs/IOCs with source references.
4. System Audit: Agent mode for comprehensive security checks and report generation.

CLGMNI represents a key evolution of terminal AI tools—moving beyond Q&A to become an operational assistant. Its secure design, cost control, and rich features make it valuable for sys admins and security pros. As Agentic AI grows, more such tools will combine LLM reasoning with domain-specific operations. CLGMNI is an excellent starting point for exploring AI in system management.