Chapter 10 · 8 minutes
Jan or LM Studio, and what next
Two good apps for the same job. Why this workshop uses Jan, when LM Studio is the better pick, and how to upgrade your model.
Goal
Know the alternative, know when to upgrade the model, and leave with a next step.
Jan and LM Studio, side by side
| Jan | LM Studio | |
|---|---|---|
| Source code | open source (Apache-2.0) | closed source |
| Price | free | free, at home and at work |
| Windows | x64 with AVX2, or ARM64 | x64 with AVX2, or ARM (Snapdragon) |
| Mac | Apple Silicon, macOS 13.6+ | Apple Silicon, macOS 14+ |
| Memory | 8 GB minimum, 16 GB recommended | 16 GB recommended |
| Chat offline | yes | yes |
| Telemetry | opt-in analytics, off by default | "does not include telemetry", per its policy |
| Update check | on by default, can be turned off | on every launch |
| Chat with documents | yes, processed locally | yes, processed locally |
| Feels like | close to ChatGPT, fewer knobs | more panels and settings |
Sources: Jan Windows, Jan Mac, Jan privacy, LM Studio requirements, LM Studio offline, LM Studio privacy, free for work.
Why Jan here: for a privacy workshop, "you can read the code" beats "trust our policy". It is also the closest to ChatGPT for a first-timer.
Pick LM Studio if: you want more control over model settings, or a policy that says no telemetry at all.
The models are the same GGUF files, so nothing you learned here is wasted.
Upgrading your model
Use the labels you noted in chapter 05.
| Your laptop | Try next | Download |
|---|---|---|
| 16 GB memory | Jan-v3.5-4B, Q4_K_M | 2.72 GB |
| NVIDIA card with 8 GB, or a 16 GB+ Mac | Qwen3.5-9B, Q4_K_M | 5.68 GB |
| Old or 8 GB laptop where Qwen3.5-2B is too slow | Llama 3.2 3B, Q4_K_M | 2.02 GB |
Bigger models land in the same data folder, so check the free space on that drive first. Llama 3.2 uses Meta's community licence, not Apache-2.0. Read it before using it for business.
Sources: Jan-v3.5-4B, Qwen3.5-9B, Llama 3.2 3B.
What next
- Chat with your own documents: the + button in a chat → Add documents or files. With a local model, files are processed on your laptop (Jan file upload).
- Developers: Jan runs an OpenAI-compatible API on
127.0.0.1:1337(Jan local API). Build on it without any cloud key. - Build a full app on local AI: the Local AI workshop.
Check it works
You can name one reason to pick Jan and one to pick LM Studio.
Exercise
Back up your threads folder to another drive or a pen drive. Then open one backed-up messages.jsonl to confirm the copy is readable.