Jan0.8.8

Jan lets you converse with AI on your computer, using local models for writing, coding, summaries, and other tasks. It operates 100% offline by default, keeping cloud use under your control when online models are unnecessary. As an open-source alternative to ChatGPT, it is actively maintained by Menlo Research and built in public.

On Linux, it can run Llama, Gemma, Qwen, and other open models directly on the device. When a hosted model is needed, Jan can connect to OpenAI, Anthropic, Gemini, Groq, GPT, Claude, and other cloud services. Custom assistants can be shaped for writing, coding, summarizing, or other roles, while conversation context and preferences carry across interactions.

Model Context Protocol support adds integrations for workflows and tooling, while the minimalist interface keeps model choice and assistant behavior accessible.

Version 0.8.8 adds backend-managed settings and credentials with OS keyring support, native web_search and web_fetch through tauri-plugin-websearch, built-in extension bundling, per-model chat-template kwargs, and an OpenAI-compatible translating gateway with Responses API support. It also persists chain-of-thought duration, guards against duplicate MCP serving, rotates API keys across providers, and includes a large batch of chat, RAG, and desktop fixes. Users can send feedback, bug reports, and feature requests through GitHub.