github.devotd.
NewAI

janhq/jan v0.8.7 — A ChatGPT alternative for running local LLMs through a GUI

A rundown of the latest release v0.8.7 of janhq/jan, the open-source ChatGPT alternative desktop app, and the local LLM execution and GUI-based model management scope described in its README, along with the fact that the available record contains neither a changelog nor license terms.

Repositoryjanhq/jan ↗

A rundown of the latest release v0.8.7 of janhq/jan, the open-source ChatGPT alternative desktop app, and the local LLM execution and GUI-based model management scope described in its README, along with the fact that the available record contains neither a changelog nor license terms.

GitHub description: Jan is an open source alternative to ChatGPT that runs 100% offline on your computer.

stars
★ 45K
Forks
3.1K
Language
Rust
License
NOASSERTION
Latest release
v0.8.7
Read original source ↗View on GitHub ↗

janhq/jan is an open-source desktop app presented as a ChatGPT alternative. The repository description states that the app runs 100% offline on your computer. The latest release is v0.8.7, and what changed in this version cannot be confirmed because the given record has no changelog.

The core of Jan is downloading and running large language models (LLM, large language model) directly. The README explains that it makes open-source AI easy to use while prioritizing control and privacy. Model management and local conversations are done in the GUI, and if needed you can optionally connect several cloud AI services.

It supports local models from HuggingFace, and provides custom assistants, an OpenAI-compatible API, and MCP (Model Context Protocol) integration. It can be run locally at any time and ships downloads for Windows, macOS and Linux. The implementation language is Rust and the GitHub star count is 45027. The repository was created on August 17, 2023.

That said, the license is only listed as NOASSERTION, so the specific usage terms are unknown. There is also no record of the v0.8.7 changelog, performance measurements or benchmarks. Among the remaining figures, the Momentum score is 3.9 (out of 10) and the other scores have no values. It cannot be confirmed whether the feature scope listed in the README matches actual behavior at the same level. So what this release improved can only be judged once the changelog appears.