Skip to main content

Case study: Aleois’ Quivr

State of Open: The UK in 2023

Phase Two “Show us the Money”

Part 2: AI Openness

Ben Ellerby, Founder

Theodo Group invests in R&D, focusing on serverless and Generative AI through Aleois. Aleois’ new Open Source web app, “Quivr”, is less than two months old, has over 50 contributors, and aids in AI solutions. Quivr’s modular backend is free to end-users and hosted at Aleois’ expense. It can be trained on various data but avoids sensitive documents. Future plans include writing documentation and direct searches. Quivr builds “separate brains” to segregate data types, addressing security concerns in the LLM space. It’s compatible with top LLMs, offering plug-and-play with limited customisation. Lastly, they emphasise human responsibility for AI outcomes and suggests a legislative precedent based on copyright, with the person creating the AI holding the copyright and responsibility for its output.

Theodo Group invests in R&D with a focus on serverless, Generative AI through its Aleois company. Two of the three founders are in London. Aleois helps companies to adopt cloud strategies and its AI product is a web app called “Quivr”, which is an Open Source work in progress, less than two months old, with 50 plus contributors, and which is currently looking for a new name. It creates an all in one solution which “Stan Girard60,’’ one of the group employees, dreamt up. The team are helping him to enact this vision, of giving anyone access to AI solutions. The whole modular back end is Open Source, almost 10,000 aApp is hosted at cost to Aleois and free to use to the end user.

Train it on your own body of knowledge, or you could give it a crm or spreadsheet, to allow the AI to train on it, then ask it for its suggestions. It will not upload sensitive documents, as there is a risk as with Gen AI, as to how secure data is. Focus on retrieving information with natural language, as if you are asking a friend. Next stage is to move to write documentation and be able to directly search it.

Building “separate brains” to ring-fence bodies of information, like sensitive and personal data. For now in the LLM space is one where companies are not ready to ensure that security which leaves the risk of data leakage which is something that companies need to react rapidly too, to ensure that their employees are not leaking data.

Quivr works off any LLM, as it is compatible with the top 5 LLMS – GPT 4 and 3.5, Vertex, Cord, and Open Source models Hugging Face and GPTforAll which was built off LLaMA. Its interoperability is its key advantage offering plug and play with limited customisation.

“I am not a fan of attributing personality to systems. Even if we did come up with a genuine artificial intelligence that thought for itself there must be human responsibility. We’ve got a legislative precedent, of course, with computer generated art works, because the person who puts into place the systems for creating the artwork actually has the copyright. And I do think that at the copyright level that should be dealt with straightforwardly. You should start with an arbitrary rule that the person who creates the AI has the copyright in and is responsible for the output.”

Iain Mitchell, Honorary KC, OpenUK

Scroll to top of the content