Bot Food raises C$1.5 million for a personal context layer—and launches a beta

A new company between the person and the model

Bot Food’s starting proposition is that a person’s useful context should not have to be rebuilt inside every AI service. The Toronto startup has acknowledged C$1.5 million in pre-seed funding from GreenSky Ventures and the launch of its RaLHF beta. The October 29 company post is a synopsis of BetaKit’s original reporting, rather than a second independent interview; the distinction matters when describing the evidence behind the financing.

BetaKit reports that the company announced the secured funding and beta on Thursday. Bot Food’s own dated post confirms the amount in Canadian dollars and the investor. This is not a US$1.5 million round, nor a report of a planned raise. The product story is a separate proposition: maintain a personal context library that can be used across AI applications.

Portable does not mean stored only on your device

On its own company page, Bot Food says the library is stored in a RaLHF account, with users deciding which AI applications can reach it. It also says it builds on MCP and open formats. Those are company descriptions of the design, not independently verified findings about security, interoperability or every supported application. The funding and a public beta do not establish those outcomes.

The account detail is important because several ideas can otherwise be blended into a single word such as control. A portable library might be hosted by a service. A user might have permission controls without every piece of information staying on a local device. The company’s stated design should not be rewritten into a stronger claim of fully offline storage or universal compatibility.

There is a practical reason to separate context from the model that uses it. A person could want an AI application to know a persistent preference without re-entering it each time. That preference is not necessarily appropriate for every application or every task. The value of a common context layer would therefore depend partly on the ability to make selective choices, not merely on accumulating more information.

The useful memory is the one you can inspect

Imagine a hypothetical user who records a work preference and later changes jobs. Carrying the old context into every new interaction could make the system more familiar while making its assumptions less appropriate. A useful memory interface would help the person see what is being carried forward, update it and understand which applications can access it. This is an illustrative evaluation question, not an observed problem in RaLHF.

Similarly, exporting a library and revoking an application’s access answer different questions. Export concerns moving information. Revocation concerns a permission relationship. Deletion concerns what remains stored and where. A product that speaks broadly about ownership should make the distinctions legible. Foundra has not tested those processes in the announced beta and does not claim that any of them is absent.

The founder challenge is thus more specific than building a smarter assistant. The context layer has to explain itself well enough that an individual can make informed choices about information. A model may benefit from additional context, but that does not make every additional disclosure desirable. The right amount of context depends on the person, the task and the permission granted.

The beta is a starting point for examination

The published financing acknowledgement establishes a commercial milestone, while the beta creates an opportunity to examine the product’s stated boundaries. A careful evaluation would ask what formats are supported, how access decisions are represented and what a user can inspect or change. It should not assume that a broad portability claim has already been demonstrated across every AI platform.

Bot Food is betting that personal context can be a distinct software layer rather than an inseparable feature of a particular AI service. Its early funding and beta are concrete news. The durability of that proposition will depend on whether portability and permission become usable choices—not just attractive language about owning one’s digital memory.