SaolixGroup
Private AI you run yourself.
Sōli AI is Saolix's applied AI programme for governed agentic AI. We are building AI agents and small language models designed to run privately: inside your own infrastructure, or on the device in your hand, so your data never has to leave.
In development Sōli AI models are not yet released. This page will carry model cards and downloads when they are.
Your data, your perimeter, your model.
Most AI today runs in someone else's cloud. For banks, regulators and anyone handling personal financial data, that is often the wrong trade.
Data stays where it belongs
A model that runs inside your perimeter, or on the device, never needs your customers' data sent to a third-party AI service in another jurisdiction.
A model you can inspect
A small model you host yourself can be versioned, tested against your own cases and switched off, like any other component you are accountable for.
Small enough to afford
Small language models are designed for everyday hardware. Useful AI should not need a data centre of specialised chips.
Starting with the phone.
Sōli AI's first model is Soli Lite: a small model designed to run on a phone and help people understand the mobile-money messages they receive every day, without those messages leaving the device.
Soli Lite
On-device intelligence for mobile-money messages. Planned: English first, then Swahili and Sheng.
About Soli Lite → AgentsCopilots in our products
Where AI already assists in Saolix products, and the line it never crosses.
See the agents → PrincipleAI advises, rules decide
Why every Saolix product keeps AI on the advisory side of the decision.
Read the article →AI agents that work inside guardrails.
Agentic AI lets software gather information, use tools and propose the next step on its own. In regulated work, the question is not whether agents can act, but who authorises them. Sōli AI agents are designed so that approved rules and people authorise every consequential action.
Raia
Eagle's optional AI agent for alert triage: it suggests a label for a screening alert, and can never close one.
AI agents → PlannedAgent Manager
One place to register, permission, monitor and switch off every AI agent in your organisation.
What's next → PlannedAgent Marketplace
A catalogue of governed AI agents for regulated work, each with its guardrails and audit built in.
What's next →Three rules for every Sōli model.
Private by design
Models are built to run where the data already is, not to pull data out to a model somewhere else.
Advisory, never deciding
In regulated work, a model can suggest and explain. Rules and people make the decision.
Published when proven
We will publish a model card, evaluation results and licence terms with every release, and not before the model has earned them.
Your next chapter.
Let’s build it together.
Explore a product, discuss a project, or find the right expertise. Tell us where you want to go. We’ll help you take the next step.