AKA Foods has moved its flagship platform from assistant to actor, launching AKA Studio Alchemy for all customers while opening a beta for AKA Atlas, a private knowledge engine designed to give food companies a verified model of their own science.

The Amstelveen-based food technology firm unveiled both products on 1 September, framing them as two halves of the same problem: R&D teams sitting on decades of formulation history, sensory data and supplier knowledge that is too scattered to use, and too slow to search, when a new project begins.

From informed to agentic

Alchemy is the third generation of AKA Studio, and the shift in ambition is explicit. Where earlier versions helped technologists find answers and organise data, Alchemy acts: creating projects, generating batches of candidate recipes, reading product labels from photographs and updating development briefs. Every action still requires sign-off.

“We built Alchemy around the loop that food technologists already run: Ideate, Formulate, Sensory, Audit,” said David Sack, CEO and co-founder of AKA Foods. “Alchemy takes on the tasks inside that loop that don’t need a scientist’s palate, the searching, the constraint checking, the compliance tracking, and the scientist stays in charge of the science.”

That loop structures the platform end to end. Teams explore directions in Ideate; Formulate generates and refines candidate recipes against layered constraints such as cost caps, nutrient targets and ingredient restrictions; sensory feedback from tasting rounds feeds directly back into the next iteration, with nutrition, cost and composition recalculating live. At any stage, a technologist can run a regulatory audit, which checks relevant requirements, flags deviations and shows which constraints held, with sourcing attached to every suggestion.

The release brings 63 improvements across the platform, including simultaneous multi-candidate recipe generation with controls for constrained versus more exploratory suggestions, brief reading from Word and Excel files, and downloadable reports covering project summaries and QA/HACCP documentation.

A knowledge engine

AKA Atlas addresses the other half of the problem: what happens to everything a company knows once the project ends. Now entering invite-only beta, Atlas builds and maintains what AKA Foods calls a private, verified world model of an organisation’s food science, connecting internal formulation history, sensory results, process learnings and supplier data with external scientific literature, patents, regulatory updates and ingredient specifications.

Queries return answers with evidence attached, sources cited and a confidence score. Where the system identifies gaps, it dispatches automated research scouts to locate and verify external information, and when that external evidence conflicts with internal data, Atlas surfaces both sides rather than resolving the discrepancy silently.

“Every food company is using AI now. Almost none of them owns any intelligence,” said Sack. “The chatbots their teams use are shared with competitors, ignorant of their business and forget everything the moment the window closes. Atlas builds the private, verified knowledge that makes AI answers actually worth trusting.”

Professor Alex Bronstein, chief scientist at AKA Foods, drew the distinction with conventional retrieval-based tools more sharply: “Most AI systems retrieve documents and summarise them. Atlas is different: it holds a model of your domain, typed, connected and scored for confidence, so answers are computed from what your organisation actually knows, with the evidence attached. Nothing trains a shared model, and your data never leaves your boundary.”

Atlas is designed to be forked by geography or category, so an ingredient company could run separate branches for, say, beverages, bakery or Brazilian regulatory requirements, each inheriting from a shared foundation and kept current by its own scouts. Integration is read-only, with nothing migrated from a customer’s existing systems. The platform supports Bring Your Own Model for companies wanting to connect their own prediction models, and Sovereign AI deployment, running on open-weight models inside a customer’s own infrastructure.

Data sovereignty as a selling point

Both products sit within AKA Foods’ private infrastructure, with the company holding ISO/IEC 27001:2022 certification and a completed SOC 2 Type II examination. Every customer environment is siloed, and the company states customer data is never shared or used to train outside AI models. For the most sensitive deployments, Atlas supports on-premise and air-gapped installation, running the full stack inside a customer’s own infrastructure with nothing leaving their boundary.

Alchemy is available immediately to all existing AKA Studio customers. Atlas is open for beta applications via aka-food.com or [email protected].