Bot Food raises $1.5M for a personal context library for AI

Each AI product you use keeps its own piece of what it knows about you. None of them share it with the next one. Bot Food, a Toronto startup, raised $1.5 million CAD to change that. BetaKit reports that the pre-seed round came from GreenSky Ventures. The company also launched a beta of its product this week.

The founders built and sold a bot management startup called Wysdom to US-based Calabrio. Ian Collins told BetaKit that the Wysdom team measured many bots. "It was almost always a context problem," he said. The idea for Bot Food came from that work. Collins said the company wants a product he describes as "one personal context library that works across every AI product."

The beta is a simple switch. A user can turn access to their personal data on or off for a given AI system. Collins said the next version will let AI systems pull context from specific apps the user chooses. The product site is ralhf.ai.

The company has seven staff. Brendan Stennett, the former CTO of ThinkData Works, has joined as CTO. The pre-seed money will go mostly to go-to-market testing, according to BetaKit.

Jeff Brunet, managing partner at GreenSky, picks startups whose tech other products will depend on. He told BetaKit the same in an email. "A universal personal context layer could become a critical piece of infrastructure," he wrote. He said it would matter for the entire AI ecosystem. He also said the deal was easy because he already trusted the founders.

Collins expects most people to accept that they need to share detailed context with AI. "A lot of people are resisting," he said. "But we're all going to give in eventually; the productivity is just going to be too great."

The product also has a risk, which Bot Food acknowledges. A central store of personal context is a target for attackers. Collins said the company wants to earn trust at the level of 1Password, and bring the same security. He cited the architecture of the product and Stennett's background in enterprise data security.

The bet matters for teams building AI features. Answer quality already depends on the context the model receives about the user. If one personal library becomes the standard store for that context, every AI product will need to read from it. Keeping a private copy will stop being useful. Teams will face that design choice before an investor's bet is proven right. The engineering work will sit on the read side of the library.

Source

• Primary source: Trevor Nichols, BetaKit.
• Company site: Bot Food (ralhf.ai).