Jan local AI model app for business: what it does and how it is set up
Jan is a desktop app for running AI models on your own computer. Learn what businesses use it for and how MILabs sets it up.
Updated 2026-01-23
What Jan is
Jan is a desktop application from Menlo Research for running large language models on a personal computer or work machine. Rather than sending every prompt to a hosted service, the app loads a model that is stored on the device and answers from it, through a chat window that will feel familiar to anyone who has used a browser-based assistant. Jan can also reach remote model providers when a business chooses that route, so the same interface can serve both local and hosted work. Local models vary in size and capability, and the hardware a business has available decides which ones run at a useful speed. Jan belongs to a wider group of tools that let teams experiment with AI in a controlled way. It is most useful where keeping text on a known machine matters as much as the answers themselves. For a small team, it offers a practical place to test prompts, draft documents and see how a model behaves before any wider rollout.
Jan in plain language
A language model is a large file of learned numerical weights. Running one means loading that file into memory and calculating a reply, token by token, from the prompt. Jan supplies the chat screen and the runtime for this on a laptop or desktop, so the calculation happens on the device in front of you.
Model size is usually described by the number of parameters. More parameters tend to give more capable answers, but they also need more memory and take longer to produce each word. To make large models fit on modest hardware, model publishers often release compressed versions, sometimes called quantised builds, which store the weights with less precision. These versions are smaller and faster, and they can lose some accuracy, particularly on long or intricate instructions.
Two other ideas matter in day-to-day use. The first is the context window, which is the amount of text the model can consider at one time, including the conversation so far and any document pasted in. The second is speed, measured in how quickly words appear on screen. Both depend on the machine as much as the model, so one model can feel responsive on one computer and sluggish on another.
What businesses use it for
- Drafting internal memos and client emails before a person reviews them
- Summarising long documents that staff would rather keep off a hosted service
- Testing prompts and comparing answers before a wider rollout
- Answering staff questions about internal procedures from a local document set
- Brainstorming product copy and marketing outlines on a laptop
- Working offline on a site or in a client office with limited connectivity
Who it suits
- Owners who want to test AI on sensitive text before committing to a platform
- Small teams with capable desktop hardware and a clear use for local models
- Technical staff who want to compare local and hosted models side by side
How MILabs sets it up
MILabs starts with the owner or team lead, agreeing which tasks Jan should cover and which should stay with existing tools. From there the team chooses the models to install, based on the hardware the business already owns and the kind of text it handles. Where a remote provider is part of the plan, MILabs connects it with the business's own accounts and keeps the keys in a place the owner controls. Permissions come next, setting which staff can open the app, which folders it can read, and who can change models or settings on shared machines. Finally, the owner and MILabs agree who looks after the installation, including model updates, hardware checks and the point of contact when something stops working.
Builds it fits
Industries it fits
Jan in practice
In a working business, a Jan setup usually sits beside existing tools rather than replacing them. A staff member opens the app on a designated machine, selects a model that suits the task, and works through a sequence of prompts: first a draft, then a revision, then a check against a source document they paste in. Someone who knows the subject reads the result, corrects it and decides what leaves the building.
The most common pattern is a small, bounded test. Staff pick one recurring task, such as turning meeting notes into a tidy summary, and run it through the local model during ordinary work. They keep a simple record of where answers were useful, where they needed heavy editing and where the model missed the point. That record shows whether a local model is good enough for the job or whether a larger machine is needed.
Over time, the work tends to settle into a few habits. Documents that should not leave the premises stay on the device. Prompts that work well get saved and shared as a short internal library. Updates follow a schedule someone owns, and the person responsible notes which model version produced each output.
Frequently asked questions
- What does a Jan setup need from me?
- A setup starts with a short conversation about the tasks you want the app to handle, an inventory of the computers that will run it, and a named person to look after it. It also helps to know which documents or systems the team already uses. If a remote provider is part of the plan, the owner agrees who holds the credentials.
- What does Jan not do?
- Jan runs models and gives you a chat interface. It does not replace a line-of-business system, manage your customer records, or decide what your staff should do with an answer. Local models can also make mistakes, so a person still reviews anything that goes to a client, a regulator or payroll. Jan works with text, and it is one part of a workflow rather than the whole workflow.
- Is Jan suitable for a small business?
- Yes, for a business that has a clear use case and a computer capable of running a model at a usable speed. Small teams often start with drafting, summarising and internal questions, because these tasks show quickly whether local models are good enough. Where hardware is limited or many people need to work at once, a hosted or shared service may fit better, and MILabs will say so early.
- Is Jan safe to use with customer data?
- Yes, for models that run on the device, because prompts are processed on that machine, which can suit sensitive records. If a remote provider is added, the text sent to that provider leaves the device under that provider's terms. MILabs limits which folders and accounts the app can reach and agrees the setup with the owner, so the business keeps control of its data.
- What are the limits of Jan?
- The main limits are hardware, model quality and maintenance. Larger models need more memory and processing power than many office computers have, and smaller models may give weaker answers on complex tasks. Models and the app need updating, and someone must schedule that work. Each model has its own licence and terms, which set out how it can be used. These limits are usually known early in the setup.
