repository·main·Indexed 12 days ago
https://github.com/janhq/janAn open-source, local-first ChatGPT replacement that allows users to run LLMs privately on their own hardware or connect to cloud-based AI providers. It includes an OpenAI-compatible API server, a core library (@janhq/core) for building extensions, and an AutoQA E2E test runner for application automation.
Jan-v2-VL is an 8B vision-language model built on Qwen3-VL-8B-Thinking. It is specifically designed for long-horizon, multi-step agentic tasks in software environments like web browsers and desktop applications. It combines language reasoning with visual perception to perform UI control, screenshot grounding, and tool calling (e.g., via BrowserMCP).
| Variant | Focus |
|---|---|
| Jan-v2-VL-low | Efficiency-oriented, lower latency |
| Jan-v2-VL-med | Balanced latency and quality |
| Jan-v2-VL-high | Deeper reasoning, higher think time |
Jan is available across multiple platforms to suit different workflows:
https://chat.jan.ai/.When choosing an alternative to ChatGPT, consider your requirements for privacy, offline access, and specific tasks. Jan is positioned as the best overall replacement because it supports both cloud and offline modes, is open-source, and supports agentic actions and MCP servers.
| ChatGPT Alternative | Offline | Key Strength | Best For |
|---|---|---|---|
| Jan | Yes | Runs Cloud + Offline, open-source | Best overall ChatGPT replacement |
| Claude | - | Strong writing and reasoning | Creative text & code |
| Gemini | - | Integrated with Google | Research tasks, image generation |
| Perplexity | - | Fast, with cited answers | Research and fact-checking |
| LM Studio | Yes | Runs open models on PC | Coding and experiments |
Jan is an open-source ChatGPT replacement designed for privacy and local control. Key capabilities include:
localhost:1337 that follows the OpenAI API specification, allowing other applications to call your local models.Jan is an open-source ChatGPT replacement that provides full control and privacy. Key features include:
localhost:1337 for use with other applications.Lucy is a compact 1.7B parameter model built on Qwen3-1.7B, specifically optimized for web search via tool calling. It is designed to work effectively with search APIs like Serper, making it suitable for resource-constrained environments where web search capabilities are needed.
Key Strengths:
The Model Context Protocol (MCP) is an open standard that allows AI models to interact with external tools and data sources through a standardized interface.
Jan acts as an MCP host, meaning it can connect to various MCP-compliant clients and servers. This allows your AI assistant to:
By using MCP, you avoid the need for custom integrations for every new tool, as any MCP-compliant tool can interface with any MCP-compliant model.
In GPU programming, a kernel is a piece of CUDA code designed to execute a sequence of operations on the GPU. Kernels are launched within a thread block grid, which executes the code across multiple Streaming Multiprocessors (SMs) on the GPU.
Benchmarking is used to quantitatively compare kernel performance across different problem sizes and to understand how hyperparameters impact execution speed. Accurate benchmarking is critical because GPU execution is complex and prone to measurement errors.
Recommended Tools for Accurate Measurement: If you require high-fidelity measurements of kernel characteristics, it is recommended to use official NVIDIA tools from the CUDA Toolkit rather than custom scripts:
ncu)Jan uses a family of specialized models optimized for autonomous agentic tasks (planning, executing, and completing tasks over long horizons):
web_search and web_fetch tools powered by tauri-plugin-websearch. These tools allow models to search the internet and retrieve page content directly within the agent loop without requiring an external MCP (Model Context Protocol) server configuration.