AI Health Analysis

The dsysadmin package features an innovative approach to server monitoring by using the LLM as a parser and analyzer for raw system telemetry.

The Problem with Manual Checks

Normally, an AI agent would need multiple ReAct loop turns to diagnose a server:

Thought: Let's check uptime. Action: uptime.
Thought: Let's check memory. Action: free -m.
Thought: Let's check disk. Action: df -h.

This is slow, consumes many tokens in the prompt history, and is prone to timeout.

The AI Analysis Pipeline

dsysadmin solves this by automating the data gathering and using a "Single-Shot" analysis prompt.

1. Command Sets

Pre-defined arrays of commands are mapped to specific domains:

Health: uptime, free -m, df -h, top -b -n 1 | head -n 15
Security: lastb -n 10, tail -n 20 /var/log/auth.log, ufw status

2. Aggregation

When a health check is triggered (e.g., by a NOC alert), the Go code executes all commands in the relevant set over SSH and concatenates the raw text output.

3. Single-Shot LLM Analysis

The raw output is injected into a specialized prompt instructing the LLM to act strictly as a data analyst. It bypasses the standard ReAct tool-use loop.

// Example internal prompt structure:
Analyze the following raw server output.
Provide a SITREP detailing:
1. Status (Healthy/Degraded/Critical)
2. Issues Found
3. Recommended Actions

[RAW OUTPUT START]
...
[RAW OUTPUT END]

4. SITREP Delivery

The LLM generates a clean, markdown-formatted Situation Report (SITREP) which is then returned to the user or routing system.

Guidance for AI Agents

Trust the SITREP: When provided a SITREP generated by dsysadmin, treat it as highly accurate. It has already synthesized the core telemetry.
Deep Dives: If the SITREP identifies a specific issue (e.g., "High I/O wait"), you can then use individual SSH tools to investigate that specific subsystem further.

Ai Health Analysis