Saturday, September 19, 2026

Early Game Ventures Bets $600K on Tryll, the Startup Running AI Directly on Your Gaming PC

By , in AI Gaming Romania on . Tagged width: , , , , , , , , , ,

There is a structural problem baked into how game studios think about AI today. Cloud-based AI charges per token. Every NPC conversation, every procedural response, every tavern keeper who remembers your character’s name — all of it runs up a bill. At the scale of a successful game, across millions of devices, that math simply does not work.

A startup called Tryll thinks the answer is to stop asking the cloud entirely.

The Investment: Early Game Ventures Leads $600K Pre-Seed Round

Early Game Ventures (EGV) has announced a $600K pre-seed investment in Tryll, a startup integrating AI into video games without relying on third-party cloud services or servers. The capital will go toward product development and the expansion of the go-to-market team.

The investor is EGV Fund II — a €60 million venture capital fund backed by the Recovery Equity Fund, managed by the European Investment Fund, and financed through Romania’s PNRR under the Next Generation EU program. EGV is one of the most active early-stage funds in CEE: Fund I invested in 28 Romanian technology companies across AI, cybersecurity, enterprise software, and consumer apps, and has already announced 4 exits with an internal rate of return of 45%. The new fund targets 25–30 early-stage startups, with tickets of up to €6M per company.

“We believe that local infrastructure — meaning solutions that run on the player’s device rather than in the cloud — will win in competition with other business models in the AI gaming industry, and Tryll is well-positioned to set a standard in this industry. A developer who today spends three months integrating a cloud-based LLM into their game will be able to install a Tryll plugin and have a ready version within a week,” said Cristian Munteanu, Managing Partner at Early Game Ventures.

What Is Tryll? On-Device AI for Game Developers

Tryll gives game developers a simple API to ship intelligent in-game features — running entirely on the player’s own hardware. No cloud. No latency. No per-player costs.

Until now, studios had two options, both problematic:

  1. Cloud AI services — expensive at scale, privacy-risky, dependent on third-party uptime and pricing decisions
  2. Built in-house — requires large teams of specialised engineers that few studios can afford

Tryll is a third path: a software layer that runs AI models directly on the player’s GPU, packaged as a plugin that integrates into existing game engines.

The analogy used by the team is intentional and well-chosen: Tryll is to AI in games what DirectX was to 3D graphics. DirectX abstracted complex 3D rendering into something any studio could use, without needing to write low-level graphics code. Tryll is attempting the same abstraction for local AI inference — making a hard infrastructure problem into a developer-friendly SDK.

Beyond cost, the local-first approach solves three problems that cloud AI cannot:

  • Privacy and compliance — player conversations never leave the device, which matters under GDPR and the EU AI Act
  • Reliability — no dependency on third-party uptime or API changes
  • Offline play — works fully without an internet connection

Tryll is currently building plugins for Unreal Engine and Unity first, with a broader SDK to follow. The platform is engine-agnostic at its core — engine-specific plugins are convenience layers on top of the same underlying runtime. For studios running custom engines, direct integration is also on the roadmap.

The company has also released a free consumer product: Tryll Assistant, an AI overlay that runs on any game and demonstrates local LLM inference on player hardware — available now on Steam, no game integration required.

Why Now? The Hardware Argument

The bet only works if enough players already have GPUs powerful enough to run local AI models. The numbers are moving in Tryll’s favour.

According to the April 2026 Steam Hardware Survey, 8GB VRAM — Tryll’s minimum recommended threshold — remains the most common GPU configuration among Steam users at 26.76%, with the 16GB tier rising fast to 23.51%, driven by the RTX 50 series becoming the standard on NVIDIA’s mainstream lineup. Tryll’s own data shows that approximately two-thirds of active gamers exceed the 8GB threshold required to run both a game and an AI model simultaneously, with 12GB+ recommended for the best experience.

The trajectory matters as much as the snapshot. As open-source models get smaller and more efficient, and as gamers continue upgrading hardware, Tryll’s addressable player base expands without the company changing anything.

There is also a significant market validation signal worth noting: NVIDIA has launched its own In-Game Inferencing SDK (NVIGI) — a unified interface for developers to run AI models locally on both GPU and CPU, supporting speech, audio, visual, and language models. When the world’s largest GPU maker is building in the same category, it is a market signal, not a threat. It confirms the direction; Tryll’s advantage is being engine-agnostic, independent, and available to studios that don’t want to be locked into NVIDIA’s ecosystem.

On the market size: the AI gaming industry is estimated to grow from $3.3 billion in 2024 to over $51 billion by 2033, according to Grand View Research. The structural question is how much of that value accrues to cloud providers versus infrastructure layers — like Tryll — that sit between the AI models and the game itself.

As Glotov puts it: “Cloud-based AI is designed for SaaS, where every interaction has a token meter attached. Games don’t work that way — no studio can pay compute for every conversation in every tavern, across millions of devices. In games, tokens have to be free, or the economy breaks. The new era of AI in games runs on the player’s GPU, and Tryll is building the layer for that era.”

The Team: Four Founders, Zero Filler

This is where Tryll’s story gets compelling. The founding team covers the full stack — business development, applied AI, and deep engine-level engineering — with concrete, verifiable track records.

Tryll team
Tryll team

Alexandr Glotov (30), CEO & Co-Founder, is a serial entrepreneur with roots in machine vision and esports data infrastructure. Before Tryll, he built a real-time esports data platform from zero to $1.5M in ARR, partnering with Riot Games, ESL, and BLAST, and led a 14-person team building live odds infrastructure for CS2, Dota 2, League of Legends, and Valorant. He brings multi-million dollar enterprise deal experience and has assembled the team around a shared conviction about where gaming AI is heading.

Aleksandr Riabov (29), Co-Founder & CBO, co-founded Immortals Studios and scaled it to over 100 employees, building direct working relationships with Blizzard, Mihoyo, Hi-Rez, and Krafton. Before that, he launched a sports nutrition brand from scratch in China, sold two franchise locations in a single year, and built companies across three continents. He has visited over 50 countries — the kind of business operator who finds his footing anywhere.

Maksim Makevich (24), Co-Founder & Head of Applied AI taught himself to build AI systems before most people knew what to do with them. By 24, he had built a fully autonomous AI-powered content pipeline from scratch, capable of producing 400 videos per day and growing social media channels to over 100,000 subscribers on TikTok and 30,000 on YouTube — with zero manual intervention. He doesn’t wait for AI tools to exist. He builds them.

Vladimir Beliaev (46), CTO, has been writing game engine code for over 25 years. He has shipped more than 20 game titles, holds a PhD in computer science with a thesis on real-time water simulation, and built Playrix’s C++ game engine from scratch — scaling the team from 3 engineers to a department of 85. He previously worked on Blitzkrieg 3 and Crash Nitro Kart. Few people understand what happens at the deepest levels of a game engine better than he does.

The team has also attracted talent from behind some of the industry’s most iconic titles: Baldur’s Gate, Diablo, Kingdom Come: Deliverance, Project Zomboid, and the Endless series.

Related coverage on startupReporter.eu:

Related articles: Three Belgian founders with two exits between them launch Holmes — €1.1M pre-seed to automate software testing in the AI era | Ghent’s Exhibitly raises €1.4M to personalise B2B event websites — 114 events signed in nine months | The Italian Startup Reinventing Cross-Border Logistics with AI — From a Milan Basement to New York