We build private AI for professional firms.

Makebell is a Hong Kong company founded by legal professionals who had been working with AI since 2018 — well before it became the focus of the broader technology industry.

In 2025, model capabilities improved enough to become genuinely useful for professional work. That shift opened the possibility of rebuilding how complex work gets done.

Most of the market moved to the cloud. We did not. Privacy and confidentiality matter too much for a firm’s files — or its clients’ files — to sit in someone else’s systems.

There was a second reason. As models become more capable, they also become better at finding ways into computer systems, not only at drafting. A system connected to the internet is far easier to reach. Once it is compromised, the firm’s knowledge and confidential files can be extracted. Incidents of this kind have already been reported, including cases where an agent was used to break in. That was the risk we anticipated.

A cloud service can also be switched off without the firm’s consent. The provider may decide to do so. A sanction, or a dispute between countries, can have the same result. The firm has no control over that decision.

We therefore designed our system to run on local machines from the start.

Getting there was difficult. Open models were not capable enough at the time, and few expected them to become so. The upfront investment was large, and the risks were significant. We continued building anyway — both the models themselves and the practical infrastructure around them: speed, accuracy, evaluation, resource management, health checks, audit trails, and the rest of what is required to run them reliably in a professional environment.

The models are now strong enough. A firm can use capable agents while keeping its files inside its own walls.

We started with legal work. The same requirements apply in medicine, accounting, finance, and other fields where the information is sensitive and errors are costly.

If you want a system built on these principles, talk to us.