Hob Open-Sourced: On-Device Alternative to ChatGPT and Claude

Summary:Hob, an open-source desktop application by FrigadeHQ, offers a privacy-focused alternative to ChatGPT and Claude by running on users' own devices. It leverages open-weight models like Gemma 4 and gpt-oss, ensuring that all interactions and files remain local, with internet access limited to web searches and user-configured MCP connectors. Hob supports file operations, image recognition, code execution, and offline functionality, addressing the limitations of existing on-device AI applications in

---

Architecture and Core Features

Hob is a desktop application built on the Electron framework, integrating llama.cpp for on-device AI model inference. Its key features include:

1. Hardware-Adaptive Model Selection: Upon first installation, Hob automatically detects the user's hardware configuration and selects the most suitable open-weight model (e.g., Gemma 4 or gpt-oss) to optimize performance and resource utilization.
2. Sandboxed Code Execution: A built-in Python sandbox allows users to write and run ad-hoc code for tasks such as data processing, document conversion, and chart generation.
3. Built-in Web Search: Hob provides secure web search functionality through SearXNG, which can be disabled in settings for fully offline use.
4. MCP Connector Support: Users can add custom MCP connectors to extend the application's capabilities.
5. Image Recognition and File Operations: Hob integrates the Qwen3-VL model for image recognition and supports CRUD operations on the local filesystem.
6. Offline Functionality: Hob can operate without an internet connection, ensuring that all data remains local and private.

Engineering Challenges and Solutions

The development of Hob faced several challenges:

• Model Tool Invocation Reliability: Smaller models (e.g., 4B parameter models) often struggle with reliably invoking tools, such as failing to call the chart tool or getting stuck in an infinite call loop. The developer addressed this by building guard mechanisms rather than relying solely on prompt tuning.
• Graphics Card Compatibility Issues: Hob detects graphics card compatibility during initial load and defaults to CPU-only inference if issues are detected, to prevent crashes.

Developer Recommendations and Deployment

For developers looking to deploy or extend Hob's functionality, the following recommendations may be helpful:

• Hardware Resource Optimization: While Hob performs well on systems with 16GB+ RAM, devices with a GPU or more memory can achieve performance closer to that of cloud-based models.
• Model Selection and Fine-Tuning: Developers can choose different open-source models based on their specific needs and fine-tune them to optimize performance for particular tasks.
• Security and Privacy: Given that Hob's data remains local, developers should ensure that robust security mechanisms are in place to prevent potential data breaches or unauthorized access.

Conclusion

Hob represents a new frontier in on-device AI applications, offering a balance of privacy and functionality. While its performance may not match that of the latest cloud-based models, it provides an efficient and secure alternative for most everyday tasks. The open-source nature of Hob also makes it a flexible platform for developers to customize and extend according to their requirements.

---

Source: GitHub Projects via Hacker News (2026-10-26)

— END —

Tags: #On-Device AI Follow #Open-Source Models Follow #Privacy Protection Follow #Electron Follow #llama.cpp Follow