AI agents browse the web in search of knowledge, but Spanish entrepreneur Elisenda Bou-Balust is convinced that your agents need better data. “The Internet isn’t the best repository of information we have as humanity,” she told The Infinite Loop.
Her latest startup, Cala AI, just came out of stealth with an API-like data offering targeted at engineering teams whose AI products need accurate and up-to-date information. This is an alternative to hallucination-prone LLMs, with the additional advantage that Cala’s structured data can be incorporated directly into code — no scraping or processing required.
The underlying concept is called knowledge graphs — which just happens to be the reason why Apple acquired her previous company Vilynx in 2020, and the topic she worked on for the Cupertino giant alongside her now co-founder and CTO, Issey Masuda Mora.
This also explains why 15-month-old Cala is already one of Europe’s hottest AI startups.
Owing to Bou’s track record and ambition, she secured funding almost right after Cala’s inception — a €7 million pre-seed round led by American VC firm Lightspeed Venture Partners, with participation from Spanish VC firms Kibo Ventures, Kfund and Masia.
The quietest big raise in Spanish tech
While Cala was only just an idea at the time, many journalists would still have loved to hear from Bou about this being Spain’s largest ever pre-seed round, as well as Lightspeed’s first investment in Spain, and to top it all, that she raised it while pregnant with her second child.
However, Bou chose to stay quiet for more than one year, only revealing the information when Cala opened up its platform. “I believe funding rounds are just a means to deliver a product, so until we actually had something to show, I didn’t want to say much,” she explained.
Bou knew it would take several months for Cala to be able to launch. While its product requires less capital than building a foundation model, it is also more complex than a wrapper. “At the infrastructure and architecture level, we’re pushing vector databases to their limits — we’re building systems that need to be 10 or 20 times the size of Wikipedia today.”
Cala AI - knowledge query (video screenshot)
Bou explained that Cala aims to be the layer that will enable AI agents to have data they can rely on to act reliably. This is particularly crucial in B2B contexts where LLMs have shown their limitations, but it also unlocks new use cases.
Cala’s bread and butter are scenarios in which verified information is key, but it is just as important that it returns that information in an easily digestible format — structured JSON with source citations. “[Agents] don't need to browse [the web], they need to navigate data silos seamlessly and get exactly what they need,” Bou wrote on LinkedIn.
Among Cala’s users, she said, some “are tracking delays at ports to see how shipments are being delayed and how that affects product prices.” But Cala is also used by startups for quantitative analysis and other financial applications, procurement, human resources, legal topics, and many more. “Everyone uses data all day long,” Bou said.
The graph that grows with every query
What’s fairly new, however, is the rise of agentic AI. But Bou is now well aware that most of its users won’t be humans. According to Cala’s FAQ, the platform can be queried via three interfaces, including MCP, making it callable from any framework with tool support.
Cala’s pricing still refers to seats but is mostly usage-based, with the understanding that its users will be both people and their agents.
While enterprise customers could bring in significant revenues, Bou acknowledges that building a whole new data layer is a capital-intensive endeavor. But even now, all she wants to think about is Cala’s roadmap. “We’re getting a lot of requests from investors, but I believe in “deliver first,” and that’s what we’re currently focused on: building the product.”
For Cala, building its product also means building its graph. Its data comes from public sources on the open web, including APIs, filtered and verified before entering the graph. The approach is not unlike OpenAI’s agentic search capability in ChatGPT, Deep Research, Bou concedes — but claims it operates “on superpowers.”
“The problem with Deep Research is that you do a Deep Research query, it gets processed, you spend a ton of credits and tokens, and then it’s discarded,” Bou claimed. Her view is that it doesn’t make sense to search the web every time your agent needs a tidbit of information. By keeping those data points available for future queries, Cala claims to be “eight times more token-efficient.”
The corollary is that Cala grows with user queries, Bou explained. “When you ask it about a topic, if it doesn’t know about that topic, it goes to look it up, processes it, and has it ready for your next query.” As a result, the company projects its knowledge graph will grow to between half a billion and a billion data entities by the end of 2026.
Making this happen will also require Cala to grow. With a headcount of 20 as of early May 2026, the Barcelona-based startup is hiring, mostly for go-to-market and growth, and plans to open an office in San Francisco. This also reflects how Bou envisions Cala: as a global company with a Spanish soul. She’s not afraid to admit it is a moonshot bet, and while the world could use more trustworthy information, it is too early to tell whether Cala will land its rocket.”
