This article was published on September 25, 2024

New UK startup raises $12M for personal AI agents with long-term memory

London-based Convergence is coming out of stealth with big ambitions


New UK startup raises $12M for personal AI agents with long-term memory

London-based Convergence has raised $12mn in pre-seed funding to further develop its personal AI agents, which can learn using long-term memory much like humans do.

Behind the startup are machine learning engineers Marvin Purtorab (CEO) and Andy Toulis (CTO).

The duo met at Shopify while working on recommender systems and AI assistants. In 2023, they joined enterprise AI startup Cohere. In April 2024, they founded Convergence, assembling a team of Google DeepMind, Meta, and OpenAI alumni.

Convergence’s Proxy AI agent has been developed for both employees and consumers.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol' founder Boris, and some questionable AI art. It's free, every week, in your inbox. Sign up now!

For employees, it can automate administrative and repetitive tasks and workflows to save time and increase productivity. For consumers, it can assist with everyday activities, such as booking trips or ordering groceries.

According to the startup, Proxy’s differentiating factor lies in its ability to acquire skills by using long-term memory and continuous learning.

“Our models have the ability to manage their own memories, decide to add things to their memories during inference time, and also delete things from their memory,” Purtorab told TNW. 

“This allows them to capture new information live and adapt and keep learning during inference.”

To enable this result, Convergence’s AI technology is built on Large Meta Learning Models (LMLMs). 

LMLMs are a new generation of  advanced machine learning models, built upon the principles of meta-learning, also known as “learning to learn.” This involves training a model on a diverse set of tasks, enabling it to extract common learning strategies and patterns.

According to Purtorab, LMLMs are based on a new subtype of the transformer architecture, which incorporates memories directly into the model. 

“They are then trained in a specific way to teach the model not only to predict the next word, but also how to manage (meaning remembering, storing, and deleting) things into and from their memory,” he explained. 

Convergence is launching its Proxy agent in beta today, with a limited number of spots available. As of today, the company is also coming out of stealth mode.

With the fresh capital, the startup will develop new models to power its Proxy assistants. London-based VC firm Balderton Capital led the funding round with participation from Salesforce Ventures and Shopify Ventures.

Update (11:50AM CEST, September 25, 2024): The article has been updated to include the comments from Marvin Purtorab. 

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with


Published
Back to top