Databiomes sets sights on moderating toxic gaming chats with first AI model

Since 2024, Databiomes has been working to build AI models capable of running locally on consumer devices, without the need for GPUs or cloud inference.

“There was a challenge there and an opportunity that just wasn’t being solved.”

The Toronto startup launched its first one today: Ctrlvox, a customizable offering designed to moderate toxic voice and chat content on video game players’ existing CPUs that is now available on Epic Games’ Fab marketplace. It is essentially a plug-in to Epic’s Unreal Engine.

Databiomes claims Ctrlvox outperforms current models like Alibaba’s Qwen3Guard when it comes to moderation. The startup says Ctrlvox can support a variety of use-cases in real-time, from early flagging to escalation and player safety reviews.

“There is a lot of toxicity out there … sometimes I just want to play games, I just want to talk to people, I just want to socialize without that worry,” Databiomes co-founder and CEO Steven Gans told BetaKit in an exclusive interview.

The CEO noted that small and large studios alike are spending a lot of money on online moderation these days but at the voice level, most of that work remains “very reactive,” with teams of human moderators that review reported remarks after the fact rather than spot them in real time. Both groups use a cheaper, local option, and Databiomes hopes to fill that gap.

Gans said the startup sees a valuable opportunity in making it possible to develop custom, “on the edge” CPU models, an opportunity that ultimately stretches far beyond just toxicity in online multiplayer gaming.

This is not Gans’s first AI company: the repeat entrepreneur previously tried his hand at building his own business, Ooblox, which focused on face identification and spotting deepfakes, and led Unreal Data, a platform within IBM.

He also worked at American chip giants AMD and Intel, and has leveraged those relationships to support Databiomes, which he co-founded with CTO Tomasz Klempka.

With Ctrlvox, Databiomes hopes to help studios seeking to detect and reduce harmful behaviour such as harassment and hate speech more effectively and affordably.

Databiomes developed Ctrlbox using its own platform, which it says leverages a proprietary inference engine and nano language models trained from scratch on customer-provided data. Gans claimed Ctrlvox cost only $60 and took seven hours to develop on that platform.

The company has raised $1.2 million CAD to date from a group that includes Antler Canada, Enigma Ventures, and Mistral Venture Partners, among others.

The goal with Ctrlbox is to begin to demonstrate that Databiomes can create accurate AI models for any use case a customer wants to enable on a CPU.

Given Gans’ love of games and some of the significant challenges studios and platforms are currently facing when it comes to moderating bad behaviour, they seemed like a good starting point for the six-person startup.

“There was a challenge there and an opportunity that just wasn’t being solved,” Gans said.

Feature image courtesy Databiomes.