MILabs
Tools

Lemonade self-hosted AI server for small business

Lemonade is an open-source local AI server. Learn what businesses use it for and how MILabs sets it up on their own hardware.

Updated 2026-06-17

What Lemonade is

Lemonade is an open-source server that runs AI language models on a computer the business controls. Its project is published by Lemonade SDK, and it can expose the models it hosts through an OpenAI-compatible interface, so applications written for that interface can send requests to a local address instead of a cloud service. The software manages model loading and handles requests from other programs. Because the models run on local hardware, the business decides which machine does the work and where the files sit. Lemonade is a serving layer rather than a chat product with its own finished screens. Teams usually pair it with an application or workflow tool that handles the front end, while the server supplies the model answers behind it. Model choice stays open, so a business can run different open-weight models for different tasks and change them later without rewriting the surrounding system.

Lemonade SDK: official site

Lemonade in plain language

A local AI server is a program that loads a language model onto hardware the business owns and answers requests from other software. Rather than sending each question to a cloud provider, the surrounding applications call an address on the local network. That arrangement changes where prompts travel, how the model is updated and who is responsible for keeping it running.

Lemonade belongs to this group of tools, and its source code is open for a technical team to read. Because its interface follows a widely used convention, programs that already support that convention can often be pointed at it without being rebuilt.

The useful distinction is between the engine and the experience. The engine receives requests, loads the right model and returns the output. It does not supply a complete experience for the people who use it. Someone still decides which questions are asked and how answers are checked.

Owners usually focus on the hardware, the model and the people who maintain both. Each model has different memory and speed requirements, so the machine and the model are chosen together. Open-weight models can be replaced as better options appear, but each replacement needs testing against the business's own tasks.

What businesses use it for

  • Answering staff questions from the company's own documents without sending them to a third-party cloud service.
  • Drafting internal summaries of meeting notes and case files on hardware the business owns.
  • Powering a private assistant that answers questions about policies and procedures.
  • Supplying the model behind workflow tools that classify incoming emails and forms.
  • Testing open-weight models against real business tasks before choosing one for wider use.
  • Keeping sensitive client records inside the business's own systems during automated processing.

Who it suits

  • Businesses that need AI answers from documents they cannot send to outside services.
  • Teams with an in-house technician who can keep a local server running.
  • Owners who want to compare open-weight models before committing to one.

How MILabs sets it up

MILabs starts by agreeing with the owner or team lead which tasks the local model should handle and which hardware will run it. We check the machine has enough memory and processing for the chosen models, then install Lemonade and select matching models. Next we connect the systems the business already uses, such as its document store, inbox or booking software, so requests reach the local server through the interface those applications expect. Access is then set by permissions: who can send requests, which folders the assistant can read, and what is logged. Last, we agree who looks after the server, including applying software updates, replacing models and watching the machine's storage. The owner keeps a written record of these decisions for whoever maintains the system later.

Builds it fits

Industries it fits

Lemonade in practice

In a working business, the local server usually sits behind something people already use. A staff member might type a question into an internal chat page, and that page forwards it to the server, which returns an answer drawn from the documents the team has allowed it to read. The person asking sees only a familiar screen.

Work of this kind tends to begin with a narrow set of tasks. A team might decide that the assistant answers questions about leave policy and equipment procedures, and leaves pricing and client disputes out. Each task then gets its own instructions, its own permitted sources and its own review step.

An accounts team could route scanned supplier invoices through a workflow tool, which sends the extracted text to the local model for classification. The result then goes to a person for approval before anything is posted to the ledger, and the model call itself stays on the business's own equipment.

Someone must be named to apply updates, watch disk space and test any replacement model before it goes live. Where there is no technician, the owner records who is responsible and what to do when the server stops answering.

good questions, straight answers

Frequently asked questions

Can I self-host Lemonade?
Yes. Lemonade is built to run on hardware the business controls, which suits self-hosting. It runs on a suitable computer or server in the office or a rented space. The owner decides which machine runs it, who can reach it over the network, and when updates are applied. MILabs plans the installation with the owner and agrees how the server is maintained. Hardware needs depend on the models chosen, so those choices come first.
Which builds work well with Lemonade?
Builds that handle private documents tend to suit Lemonade best, because the model can run where those files already sit. Examples include a private document question-and-answer server, a knowledge base chatbot for staff and a policy compliance assistant. Meeting notes and case notes can be summarised locally when the records are sensitive. Very large models or live web search may need a different arrangement, so each build is assessed against the hardware and data.
What is Lemonade?
Lemonade is an open-source server from the Lemonade SDK project that runs AI language models on local hardware. Applications send it requests through an OpenAI-compatible interface, and it returns the model's responses. It is a serving layer, so it handles loading models and answering requests, while businesses usually build the user-facing part in a separate tool. Typical partners are workflow tools and chatbot front ends. The source code and documentation are published on GitHub.
How does MILabs set up Lemonade?
MILabs begins with the owner or team lead, agreeing which tasks the local model should handle and which machine will run it. The server is installed, suitable models are selected for those tasks, and the systems the business already uses are connected to it. Permissions are set so the right people and applications can send requests. The business agrees who maintains the server and its models. Scope and cost are confirmed through the enquiry form.
Where does Lemonade store data?
Lemonade runs on the machine the business chooses, so prompts and responses are processed on that machine. Model files sit on that machine's storage. Any logs or saved records are kept by the connected application, and where they go is agreed with the owner during setup. Backups, access controls and retention periods are also agreed at that point, so the business knows exactly where its records live.
let's make something happen

Workflow automation: how we can help

Choose how you want to work on this with MILabs. Both options start with an enquiry.

Discovery is free, no obligation from the first chat.

Let us build it for you

We build it and look after it.

Build it yourself, we'll help

You build it with our direction, teaching and a weekly check-in.