LuminidsAI: Solo Founder's CRI Gives Game Characters Memory
In a small forest clearing inside a video game, a creature named Astra is still searching for her friend Quill. She saw him caught in a storm hours ago and hasn't seen him since. What she doesn't know is that Quill found shelter almost immediately and has been perfectly safe the entire time. The game knows this. Astra doesn't. And that gap, deliberately, is the whole point.
It's a small moment inside Luminids, an Early Access colony-sim built by a solo developer in Cambridge. But it's also a working demonstration of something Nick Jacob has spent the better part of a year building: a system that lets AI-driven characters hold onto their own version of events, separate from what the game world itself knows, and separate from whichever AI model happens to be generating their dialogue. He calls it CRI, short for Contextual Relational Intelligence, and he's now raising £1.5 million to turn it into a product other studios can use.
Reviewed by Imran Khan Pathan, Editor at AI Tech Safar. Facts cross-checked against official filings and verified sources.
Quick Summary
- Jacob launched LuminidsAI, the company behind CRI, roughly two weeks ago.
- CRI tracks what a game character has actually seen, remembered and believed, independent of whichever AI model happens to be generating that character's dialogue.
- It's already running inside Luminids, his own Early Access colony-sim on Steam, which has real numbers behind it: 335K+ Steam store visits, and by Jacob's own tracking, 30M+ social views and 30K+ followers across YouTube, Instagram and TikTok.
- His company, NJX Technologies Ltd, is a genuine UK-registered company. We checked Companies House directly: No. 16405274, incorporated 24 April 2025, listed as active.
- He's had early conversations with investors, though nothing is signed yet.
This piece is based on direct correspondence with Nick Jacob, his pitch deck, and independent checks of his Steam listing, LinkedIn history and UK company registration.
Official Luminids artwork. Source: LuminidsAIThe Problem Jacob Set Out to Solve
Most AI-driven game characters are built the same way: a language model generates dialogue in response to what's happening on screen. It's a system optimised for conversation, and it's gotten remarkably good at it. What it hasn't solved, in Jacob's view, is continuity. If a studio swaps out the model running underneath a character, or upgrades to a newer version, what happens to everything that character supposedly knew, felt or remembered? Does it simply reset?
That question is what CRI is built to answer. Rather than storing a character's memory, beliefs and relationships inside the model itself, CRI keeps that information in a separate layer that sits between the game world and whatever AI is generating the character's speech. The model can change. The character's history doesn't have to.
Astra and Quill are how Jacob demonstrates it in practice. Both live through the same storm inside Luminids. Quill gets caught outside, then reaches shelter. Astra sees the first part but not the second, so her belief about the situation never updates, and she acts on the information she actually has rather than the information the game world has. It's a small example, but it points at a bigger idea: two characters occupying the same world can legitimately know different things, and CRI is the system that keeps track of who knows what.
Behind that example is a more structured pipeline. Jacob's deck breaks every experience a character has into nine stages: saw, remembered, believed, weighed, wanted, chose, acted, learned, changed. Only information a character actually receives becomes part of its point of view. What it believes can shift as new evidence comes in, and its actions follow from that belief rather than from a line of dialogue written in advance. It's a deliberately granular attempt to model how a single event turns into a lasting change in behaviour.
The part worth paying attention to is how it's tested. Jacob's team can hold a scene's starting conditions exactly constant, remove a single belief from a character's internal state, replay the identical event, and watch the resulting reaction simply not happen. If deleting one belief deletes the behaviour that depended on it, that's a reasonable indication the system is doing genuine causal work, rather than just keeping a log of what occurred.
Built Inside a Real Game, Not a Demo
What separates CRI from a lot of AI-character pitches is that it wasn't built as a standalone demo first. It grew out of an actual production problem inside Luminids, the cozy colony-building sim Jacob has been developing on his own in Godot. Currently in Early Access on Steam, it has players growing a settlement of small glowing creatures across forests, deserts and frozen landscapes, watching relationships between those creatures develop over dozens of hours of play.
The game has found an audience largely without help from Steam's own discovery algorithms: 335K+ store page visits from more than 50 countries, with the strongest interest coming from Brazil, the US and Japan. Jacob's own devlogs, dating back to April, chart the community's growth from zero to roughly 19,000 followers and 15 million interactions in under four months, a trajectory that lines up with the larger figures in his current pitch deck. Around 1,000 playtesters have logged real hours with the game.
Both the game's progress and CRI's early development show up regularly on Luminids' YouTube channel, and on the Luminids website.
A One-Person Company, for Now
Strip away the pitch deck's language, and LuminidsAI is currently a company of one. Jacob's own background is in software engineering and cloud platforms, with stints at Citrix, Centrica/British Gas and Merck Animal Health, along with work on UK government technology programmes, before he left to build Luminids full time. His history is documented on LinkedIn.
The team the deck describes — a platform engineer, a developer tools engineer, a cognitive systems engineer, a solutions engineer — doesn't exist yet. Those are roles Jacob intends to fill if the round closes. Right now, one person is simultaneously building a commercial game and the memory infrastructure underneath it, a fact worth holding onto when reading the timelines further down this piece.
What does check out is the paperwork. NJX Technologies Ltd, the entity behind Luminids Studio, LuminidsAI and Jacob's related intellectual property, is an active UK company, incorporated 24 April 2025 and confirmed directly against the public Companies House register.
What the £1.5 Million Actually Buys
The round is structured as a UK pre-seed raise, SEIS/EIS eligible, aiming to cover roughly 18 months of runway:
| Allocation | Amount | What It Covers |
|---|---|---|
| 62% | £930k | Hiring the core platform team |
| 13% | £195k | Product and infrastructure |
| 10% | £150k | Customer validation and go-to-market |
| 10% | £150k | Legal, IP, and operations |
| 5% | £75k | Contingency |
The intent is to extract CRI from Luminids entirely and rebuild it as a licensable toolkit: CRI Studio for authoring and inspecting a character's beliefs, CRI Bench for the replay-testing process described earlier, and CRI Trace for keeping a record of what a character experienced, what it knew, and why it acted the way it did. The validation path Jacob has laid out is specific rather than aspirational — 20+ customer-discovery conversations, narrowing to 5 design partners, then 3 external pilots, then the first paid deployments — a staged, fairly conservative plan rather than a growth chart pointing straight up.
Asked how those investor conversations were going, Jacob confirmed they were underway but asked that the specifics stay vague for now, since nothing has been finalized.
Where CRI Fits Among the Bigger Players
Jacob is careful not to position CRI as a competitor to the major AI labs or the existing character-AI infrastructure companies. His pitch draws a clear line: OpenAI, Anthropic and Google handle language and reasoning; companies like Inworld ($125M+ raised) and Artificial Agency ($16M funded) handle real-time behaviour and perception. CRI, in his framing, sits underneath all of it, as the layer that stays constant even as the model generating a character's dialogue changes around it.
He points to broader signals that the industry is already moving in this direction, citing Nvidia's ACE platform and Krafton's PUBG Ally as evidence that publishers are investing in characters built around persistent state and memory, not just more fluent dialogue. Whether that translates into real demand for a standalone memory layer, built by a single founder in Cambridge, is essentially the question this funding round exists to answer.
What Comes Next
Jacob is targeting a full launch for Luminids around February 2026, with CRI reaching external pilot studios sometime in Q2 2027, assuming the round closes on schedule. Both timelines rest on the same premise: that a solo founder can successfully turn into a funded, staffed company in the meantime. That's the risk sitting underneath everything else in this story.
For updates on LuminidsAI and Luminids, Jacob can be followed on LinkedIn, with the game itself on YouTube, or via luminids.ai and luminids.com directly.
FAQ
Is LuminidsAI a real company?
Yes. NJX Technologies Ltd is registered with UK Companies House (No. 16405274), incorporated 24 April 2025, and listed as active on the public register.
Is Luminids an actual playable game?
Yes. It's live on Steam in Early Access right now, not simply a concept in a deck.
How big is the LuminidsAI team?
Just Jacob, currently. The roles in his pitch deck are planned hires contingent on the raise, not existing staff.
What does CRI stand for?
Contextual Relational Intelligence — a persistent memory and identity layer that sits between a game's world and whatever AI model is generating a character's dialogue.
Has any funding actually closed?
Not yet. Jacob describes the investor conversations as early, with nothing signed as of this writing.
Related Reading on AI Tech Safar
- Agentic AI Explained: What It Is, How It Works, and Why 2026 Is Different
- Frontier AI Architecture: Mastering the GPT-6 Astra Ecosystem
Awesome!!
ReplyDeleteThank you so much 🙏
ReplyDelete