CrashAnalizerSuite

CrashAnalizerSuite analyzes Windows crash dumps (.dmp) and supports prompt-based Q&A over crash data using a local LLM.

UI Preview

Architecture

flowchart TD

    subgraph CrashGenerator["CrashGenerator.Cpp (Native Crash Producer)"]
        CG["Crash Generator App<br/>C++ / Win32"]
        CG --> DMPFILE["CrashGenerator_*.dmp"]
    end

    subgraph App["CrashAnalyzer.WpfApp"]
        UI[UI Layer]
        SVC[WpfCrashAnalysisService / Shared service]
        UI --> SVC
    end

    subgraph Core["CrashAnalyzer.Core"]
        DP["DmpParser - CLRMD"]
        RP["RagCrashAnalyzer"]
        EC["EmbeddingClient"]
        LC["LlmClient"]
        VS["VectorStore (in-memory)"]
    end

    subgraph LLM["Ollama (Local Host)"]
        O1[llama3.1]
        O2[nomic-embed-text]
    end

    DMPFILE --> SVC
    SVC --> DP
    SVC --> RP
    RP --> EC
    RP --> LC
    RP --> VS

    EC --> O2
    LC --> O1

Loading

LLM configuration used

From crashanalyzer.settings.json:

• LLM host/base URL: http://localhost:11434
• LLM model: llama3.1
• Embedding model: nomic-embed-text
• Timeout: 300 seconds

Endpoints used by the app

• GET /api/tags - health check / reachability
• POST /api/generate - text generation for crash Q&A
• POST /api/embeddings - embedding generation for RAG retrieval

How the app contacts the LLM

The app sends HTTP requests to Ollama at LlmBaseUrl.

Minimal C# endpoint connection examples:

using var http = new HttpClient { BaseAddress = new Uri(settings.LlmBaseUrl) };
using var health = await http.GetAsync("/api/tags");

Installation / setup

1. Install prerequisites:

   • Visual Studio 2026 (18.x) with:
     • .NET desktop workload
     • Desktop development with C++
   • .NET 10 SDK
   • Ollama

2. Pull required models:

   ollama pull llama3.1
   ollama pull nomic-embed-text

3. Start Ollama service:

   ollama serve

4. Verify/update crashanalyzer.settings.json in app output (or project) to match your environment:

   • LlmBaseUrl
   • LlmModel
   • EmbeddingModel
   • SymbolizerPath

Recreate and run the solution

1. Open CrashAnalizerSuite.slnx in Visual Studio.
2. Build solution (Build > Build Solution).
3. Build and run CrashGenerator.Cpp in Debug|x64 to generate test artifacts:
   • .pdb file from the Debug build
   • .dmp file generated after intentional crash (CrashGenerator_*.dmp in output folder)
4. Analyze the dump using the app:
   • Open .dmp
   • Open matching .pdb
   • Ask questions in the prompt area

Notes

• RAG vector storage is in-memory and resets between sessions.
• The default host is local Ollama (localhost:11434); no cloud endpoint is required.