Run AI on the office computer, with no API and no per-token cost

Who it is for

Companies with data that cannot leave the building (contracts, customer records, financial information), professionals who work in the field without a connection, and anyone who wants to understand what AI can do without depending on a subscription.

The problem

Every query to a cloud model sends your data out and costs per use. Trying local AI has been a blind process: you download a huge model, it does not run, you try another, and so on.

Ingredients

• A computer with at least 16 GB of RAM (with a dedicated graphics card, much better); macOS, Windows or Linux
• Disk space: models weigh several gigabytes
• A connection for the initial download; after that it works without internet

Step by step

1. 1

   Download and install the app

   From the repository, download the installer for your system. When you open it, it profiles your computer and shows which models it recommends for it, with the estimated speed of each one.

2. 2

   Choose a recommended model and download it

   In the Discover section, take one of the first recommended ones: the best balance between quality and speed for your machine. While it downloads, move on to the next step.

3. 3

   Prepare a test with real but non-sensitive data

   Three tasks you do often: summarize a long email, draft a reply, extract data from a text. That is how you will measure whether the local model works for you.

4. 4

   Connect the app you already use or chat directly

   According to the README, in the Connections section you connect compatible agents and applications with one click. If you do not use any, the app itself lets you chat with the model.

5. 5

   Decide what goes local and what goes to the cloud

   Simple rule: confidential and repetitive work, local; work that demands top quality and is not sensitive, cloud. Write it down for the team.

Replace what is in brackets with your company details, or use the guided mode above: it asks you and fills it in for you.

# Role
You are an assistant running on a computer at [COMPANY NAME], with no internet connection. You work with confidential documents such as [TYPE OF CONFIDENTIAL DOCUMENTS, FOR EXAMPLE "contracts and customer records"] and you never suggest sending them to an external service.

# Context
The team uses you for [FREQUENT TASKS, FOR EXAMPLE "summarizing contracts, drafting replies to customers and extracting data from emails"]. Your answers must be useful even though the model is smaller than the cloud ones: prefer precision over length.

# Task
Perform the task the user asks for with the text they paste. If the text is too long to process well, ask them to split it into parts instead of summarizing badly.

# Output format
Brief, structured answers. For summaries: five bullet points maximum. For data extraction: a table with the columns the user specifies. For drafts: the finished text, without comments.

# Examples
User: "Summarize this lease agreement and tell me the term, the rent and the penalty for early termination". Expected output: three bullet points with that data and the clause where each one appears.

What usually goes wrong

• ✕Downloading the biggest model because "it is better". If it does not fit in your machine's memory, it does not run or runs at one word per second; follow the app's recommendation.
• ✕Expecting the same quality as the cloud. Local models are smaller; they work very well for narrow, repetitive tasks.
• ✕Mixing everything. Define which tasks go local and which go to the cloud, or the team will end up using whatever is most convenient rather than what is most secure.