Capability · Framework — fine-tuning

Jan

Jan is an open-source desktop app from Homebrew Computer Co. that lets you run local LLMs with a polished ChatGPT-style UI, no account required. It ships its own inference engine (Cortex, built on llama.cpp and TensorRT-LLM), exposes an OpenAI-compatible server on localhost, and supports extensions for remote providers like OpenAI, Anthropic, Groq. Popular with developers who want the Ollama back-end experience plus a real GUI.

Framework facts

Category: fine-tuning
Language: TypeScript / Rust
License: AGPL-3.0
Repository: https://github.com/menloresearch/jan

Install

# Download from https://jan.ai or
brew install --cask jan

Quickstart

# Launch Jan; Settings → Advanced → API Server → Start
curl http://localhost:1337/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "llama3.2-3b-instruct", "messages": [{"role":"user","content":"hi"}]}'

Alternatives

LM Studio — closed-source but polished
Ollama — CLI-first
GPT4All — open-source desktop

Frequently asked questions

Is Jan really offline?

By default, yes — Jan downloads models from Hugging Face on demand, then runs inference entirely locally. Telemetry is off by default and remote providers are opt-in per model.

What's the Cortex engine?

Cortex is Jan's swappable inference engine; it wraps llama.cpp for GGUF models and can also drive TensorRT-LLM engines for NVIDIA GPUs.

Sources

Jan docs — accessed 2026-04-20
Jan GitHub — accessed 2026-04-20