AI research and development lab Tzafon has closed a $9.7m pre-seed round — Sweden’s largest ever, the company claims.
Tzafon is building foundational models that interact with computers like humans — clicking, scrolling and inputting text. Its infrastructure can support multiple, autonomous AI agents that work together to solve complex tasks and interact with operating systems, applications and web browsers.
The company originally secured $4.4m led by Streamlined earlier this year, with participation from Kakao VC, Oliver Jung and angels from openAI and xAI. HV Capital then led an extension, bringing Tzafon’s total funding to $9.7m.
Tzafon will use part of the funding to launch its new product, Lightcone, an agent that can act on a user’s behalf across apps and platforms, in a few week’s time.
One use case for Lightcone could be helping people input data into an Excel spreadsheet, Tzafon cofounder and CEO Noah Löfquist tells Sifted: “But you can also have the model jump between different applications, give it small tasks like post something on social media, or more monotonous stuff like ‘get my receipts and put them in my QBS [Quickbooks]’.”
Lightcone is downloadable and is designed to run locally on a computer like an app. “It should feel like it’s a part of your computer and not a part of the browser itself,” says Löfquist.
Tzafon isn’t alone in building autonomous agents that can act on a human’s behalf, and there are several companies (including deep-pocketed BigTechs like OpenAI and Google) tackling different aspects of the problem.
For example, Stockholm-based Strawberry Browser is building AI “companions” that learn and imitate a user’s workflow, performing actions autonomously on websites, like research or content writing.
Fern Labs, a London-based startup founded by former Palantir engineers, is building large-scale networks of AI agents that collaborate to solve tasks like creating and testing software or data analysis.
Löfquist says Tzafon stands out from its startup competitors as, due to a recent partnership it announced with Google Cloud, it has a “pretty insane amount of compute.”
“That means we can train and run our own models. We can design everything from scratch and that also means that we have full control of the end user experience,” says Löfquist. “Depending on what we want that user experience to be, we can fully shape that without having to be reliant on anyone else’s technology.”
Tzafon currently has 15 employees — some of whom previously worked at companies such as DeepMind, Palantir and Character AI — based across Stockholm, Tel Aviv and San Francisco.
It plans to hire a few more AI researchers this year, but like many AI-native companies it’s careful to not make its organisation too big.
“We might level out at 25-30 people” says Löfquist.
“I think we want to keep being a small team, because one of the risks you get into as an AI Lab is that it’s very easy to fall into a product-focused company and that kind of destroys a lot of the opportunities you have to build even greater products, which is ironic in some ways.”
Read the orginal article: https://sifted.eu/articles/tzafon-closes-pre-seed-round/