AI Assistant

The AI Assistant provides intelligent anomaly detection and diagnostic analysis for DDS/RTPS traffic. It combines rule-based pattern detection with optional LLM-powered analysis.

Overview

Access the AI Assistant via View > AI Assistant or press Ctrl+A.

The assistant offers two modes:

Pattern Detectors (15 Rules)

Critical Severity

High Severity

Medium Severity

Low Severity

Detection Results

Each detection includes:

pub struct Anomaly {
    pub pattern: String,        // e.g., "CascadeFailure"
    pub description: String,    // Human-readable explanation
    pub confidence: f64,        // 0.0 to 1.0
    pub severity: Severity,     // Critical, High, Medium, Low
    pub suggested_fix: Option<String>,
    pub timestamp_ns: u64,
}

LLM Backends (v0.3+)

For deeper analysis, the AI Assistant can query LLM backends.

Supported Backends

Ollama Setup (Recommended)

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

Pull a model
ollama pull llama3.2

Verify
ollama list

Configure in ~/.config/hdds-viewer/config.toml:

[ai]
backend = "ollama"
model = "llama3.2"
base_url = "http://localhost:11434"

Claude Setup

[ai]
backend = "claude"
model = "claude-3-5-sonnet-20241022"
api_key_env = "ANTHROPIC_API_KEY"  # Read from environment

OpenAI Setup

[ai]
backend = "openai"
model = "gpt-4o"
api_key_env = "OPENAI_API_KEY"

RAG (Retrieval-Augmented Generation)

The AI Assistant uses RAG to provide context-aware responses:

1. Vendor Knowledge Base - Indexed documentation from HDDS, FastDDS, RTI Connext

2. Code Context - AST-indexed source code via tree-sitter

3. Session History - Recent analysis results and patterns

RAG Store

[ai.rag]
enabled = true
store_path = "~/.config/hdds-viewer/rag.sqlite"
embedding_model = "nomic-embed-text"  # For Ollama

Prompt System

The AI Assistant uses Handlebars templates for unbiased prompts:

You are a DDS/RTPS protocol expert analyzing network traffic.

Context:
Capture file: {{capture_file}}
Duration: {{duration_seconds}}s
Total samples: {{total_samples}}

Detected anomalies:
{{#each anomalies}}
{{this.pattern}} ({{this.severity}}): {{this.description}}
{{/each}}


Analyze the root cause and provide actionable recommendations.

Templates are located in:

• Linux: ~/.config/hdds-viewer/prompts/
• macOS: ~/Library/Application Support/HDDS Viewer/prompts/

CLI Usage

# Run pattern detection only
hdds-viewer --analyze capture.hddscap

With LLM analysis (requires backend)
hdds-viewer --analyze capture.hddscap --ai

JSON output for CI/CD
hdds-viewer --analyze capture.hddscap --ai --format json

Causal Analysis

The AI Assistant builds a causal graph to identify root causes:

[Upstream Failure] → [Cascade Event] → [Downstream Impact]
         ↓
   [Root Cause Identified]

Access via Tools > Causal Graph or in the AI Assistant panel.

Response Caching

To reduce API calls, responses are cached:

[ai.cache]
enabled = true
ttl_seconds = 3600  # 1 hour
max_entries = 1000

Cache location:

• Linux: ~/.cache/hdds-viewer/ai-cache/

Best Practices

Offline-First

The AI Assistant works 100% offline with pattern detection. LLM backends are optional.

# Fully air-gapped usage
hdds-viewer --analyze capture.hddscap --no-network

Privacy

When using cloud backends (Claude, OpenAI):

• Only anonymized statistics are sent
• Payload contents are never transmitted
• Enable dry_run to preview prompts without sending

[ai]
dry_run = true  # Preview prompts only

Exit Codes