exo AI cluster software: what it is and how it is set up
exo runs one AI model across several local computers. Learn what it does and how MILabs sets it up for a business.
Updated 2026-06-09
What exo is
exo is open-source software from exo labs that links several ordinary computers on a local network so they can run an AI model together. Rather than sending every request to a remote provider, a business can spread a model across machines it already owns, such as a few laptops or desktops. The project and its installation notes are kept in a public GitHub repository. Machines on the same network find each other and share the work of running the model. Software that already speaks the common chat completion format used by many AI tools can be pointed at the cluster instead of a cloud service. In neutral terms, exo is a model provider that runs inside the business's own walls. For an owner, the appeal is control. Prompts, client files and internal notes can stay on premises, and the main running cost is hardware the business has already bought. The trade-off is a more hands-on setup than a hosted service, and supported models change between releases.
exo in plain language
Many AI tools run on a provider's servers, which decides where the computation happens and how long records are kept. A local model reverses that arrangement. The model is loaded onto hardware in the building, so questions and answers travel only as far as the office network. Records can then stay on hardware the business controls. The practical limit is memory, since a capable model needs a large amount of fast memory. Splitting the work across several machines lets a team pool that memory.
Treat the setup as a small internal service rather than a single app. Someone decides which machines join, keeps their software current and watches for one that drops off the network. The model is only one part, so the network, the interface staff use and the record of changes each have an owner. Model choice matters too. Smaller models respond quickly on modest hardware, and answers are best checked against real questions from the business's own work.
What businesses use it for
- Keeping client files and internal documents searchable without uploading them to a cloud service.
- Drafting first versions of client briefs and internal memos from notes the team has already written.
- Answering staff questions about policies, procedures and past project records.
- Summarising meeting notes into a list of actions and owners.
- Testing AI features on hardware the business owns before deciding whether to pay for hosted tools.
Who it suits
- Firms that handle client records and want those files to stay on premises.
- Teams with spare capable computers and an IT contact willing to maintain them.
- Owners who want to test local AI before committing to a hosted subscription.
How MILabs sets it up
Setup starts with a conversation with the owner or team lead about which work stays in-house and which can use hosted tools. Together we choose the machines to include, check that they can reach each other on the office network, and decide which model size fits the hardware. We then connect exo to the systems the business already uses, such as the shared drive, the document store or the chat tool the team reads each day, so requests come from familiar places. Access is limited to named users, and each staff member sees only the material their role allows. Finally we agree in writing who looks after the cluster, who installs updates and who responds when a machine drops off the network.
Builds it fits
Industries it fits
exo in practice
In a working office, the cluster usually sits in a locked comms cupboard or quiet corner, with its machines on the same switch as the staff computers. Most staff never touch it directly. They reach it through a familiar surface, such as a search box in the shared drive or a prompt window in the team's chat tool. A paralegal might ask which clause in a set of precedents covers a particular indemnity, and the answer can point back to its source page so the reader can check it. A bookkeeper might paste messy notes and ask for a tidy list of open items.
The routine matters as much as the software. Someone on the team reviews a sample of answers each week against the source documents and logs any that were wrong. Updates go on when the office is quiet, and machines restart in a set order so the model returns cleanly. If a machine drops off, staff use the hosted tool they already know until it returns, and everyone knows that path in advance. The list of approved tasks grows slowly, as confidence builds.
Frequently asked questions
- How do I know whether exo is paying off?
- Start by writing down the task exo will handle and how long it takes today. Then compare those figures over a set period. Because exo runs on machines the business owns, the main test is whether the team uses it for real work and whether outputs need less correction. MILabs can set up a simple log, with the figures drawn from the business's own records.
- Which industries use exo?
- exo suits businesses that handle sensitive records and already have capable computers, such as legal services, accounting and bookkeeping, financial planning, aged care and disability services, and mental health counselling. These sectors carry strict duties about where client information is stored, which makes keeping model work inside the office attractive. Other firms fit if their staff can maintain local hardware. Fit comes down to the business's own obligations and the hardware it already runs.
- What do businesses use exo for?
- Common uses include searching internal documents, drafting first versions of briefs and notes, answering staff questions about policies, and turning meeting notes into action lists. The shared factor is that the source material stays on hardware the business controls. MILabs usually begins with one narrow task, such as a document question-and-answer tool, and extends it only after the team has tested the results.
- Does exo need technical skills to run?
- Yes, some technical comfort helps. Installing and updating exo follows the instructions in its GitHub documentation, and a cluster needs machines configured to see each other on a network. A full engineering team is not required, but someone must follow installation notes, restart a machine and report a fault. MILabs handles the initial setup with the owner and agrees who maintains it afterwards, so the day-to-day person is not left to work it out alone.
- Can exo connect to the software I already use?
- Yes, software that uses the common chat completion format can be pointed at exo, because exo offers an interface of that kind. Whether a given product connects depends on its own settings and on the network the cluster sits on. MILabs checks those details during setup and links the systems the business already relies on, such as its shared drive or chat tool. Anything that cannot be connected cleanly is named in the handover.
