Video summary

Local Jarvis Installation -- Step by Step Guide

Main summary

Key takeaways

Technology

Summary of the Video: Local Jarvis Installation (Step-by-Step Guide)

  • Goal: Install and run a “Jarvis Mark 40” AI assistant entirely locally on your computer—without needing internet access.
  • Security & privacy claim: The assistant is described as secure and free to use because it:
    • Does not send API requests to external providers like ChatGPT, cloud services, or Gemini
    • Does not send important data to remote data centers
    • Keeps processing on the local machine
  • What it can do (as shown):
    • Ask for and play relaxing/rain video content from YouTube (when internet is available)
    • Understand multiple languages (e.g., Japanese, Turkish, French, English) and switch based on user input
    • Provide cooking advice offline (pure local response generation)
    • Attempt WhatsApp messaging that fails while offline—used to demonstrate the difference between local capabilities vs. online services
  • Core implementation idea:
    • Uses local speech-to-text and a local language model runtime
    • Whisper is used for transcription, with local models selectable
    • An on-device LLM runtime is used via Ollama (other options are mentioned, but Ollama is the focus)

Step-by-Step Setup Guide (As Described)

  1. Download code from GitHub

    • Open the GitHub link from the video description
    • Download the project as a .zip, then extract it

  2. Edit the project workspace

    • Open the extracted folder in VS Code
    • Delete a specific folder (no longer needed)
    • Avoid running setup.py, because the provided version already includes the required installs (models/UI components)

  3. Run the app

    • Run main.py in VS Code using F5

  4. Select a local speech model

    • Choose Whisper
    • Select “base” for a faster option (the video mentions speed and approximately higher/accuracy tradeoffs)
    • Set language to Auto unless you want a fixed single language

  5. Choose the local LLM provider

    • Select Ollama (primary in this tutorial)
    • Note: other providers/runtimes may exist, but this guide demonstrates Ollama

  6. Install Ollama

    • Download and install Ollama for your OS (example uses Windows)

  7. Pull/install a local LLM model

    • Use Ollama’s command to pull a model by name
    • Example model mentioned: “Cuban model” (possibly mis-transcribed)
    • Hardware mention: RTX 3060, which the creator claims is a good fit

  8. Select/copy the model name

    • Use ollama list to see available models
    • Update the code configuration to use the selected model

  9. Pick a voice/TTS option for offline

    • Use a local voice option: “Kokoro”, described as running on your computer (enables offline use)
    • Mentions other voice approaches (e.g., “11 apps or HTTS,” likely transcription errors) may require internet, making them less suitable for offline mode

  10. Run Jarvis fully

    • After setup, run the application and verify behavior

Offline Testing Results (Key Demonstrations)

  • The creator disconnects internet and tests:
    • Jarvis can still understand spoken commands and respond (local Whisper + local LLM)
    • YouTube-related requests can still trigger UI/content search attempts, but external actions may be limited without connectivity
    • WhatsApp messaging fails offline, illustrating:
      • Local reasoning/answers work offline
      • Online services/actions require network access
  • The video also claims multi-language speech works correctly even offline

Troubleshooting / Install Fixes Mentioned

  • If you encounter a Python library error (example described as: “song device could not result,” likely mis-transcribed):
    • Run pip install ... for the missing package
    • The creator specifically suggests installing sounddevice via pip
  • They recommend checking previous videos for additional installation troubleshooting

Product / Model Notes & Ongoing Updates

  • Mentions “Mark 40” specifically and suggests improvements will arrive week by week
  • Encourages viewers to provide feedback/requests in the comments
  • Claims that when they create a new model, they’ll make it public via community posts/links

Main Speaker / Sources

  • Speaker: The YouTube creator (unnamed in subtitles) providing the step-by-step guide for “Jarvis Mark 40”
  • Sources referenced:
    • The GitHub repository linked in the video description (tutorial codebase)
    • The Ollama website (local LLM runtime)
    • Whisper (local speech-to-text)
    • YouTube (demo content source)

Original video