Skip to content
A floating island in the evening: eight small luminids around a campfire, with huts, lanterns, a pond and a berry bush by an old oak.

Persistent cognition infrastructure for game characters

Build characters that remember, change and remain coherent across long-running worlds.

LuminidsAI is building CRI Platform: a reusable runtime and developer toolchain for persistent character identity, memory, beliefs, relationships and expectations.

Built first inside Luminids, our own game, CRI is being productised for game teams where character continuity is already a production problem.

Explore CRI Platform See the recorded proof

Each luminid acts on what it has seen, not on what is true. Move the berries, or open one to see what it believes.

With JavaScript on, this colony runs live. You can move the berries, start a chase or bring a storm, and see which luminids notice.

A simplified model made for this page. See it recorded in the game

01 The problem

Characters can talk. Keeping them coherent over time is still bespoke.

Studios building persistent characters still have to connect memory, relationships, save state, game events, model boundaries, authoring, debugging and behavioural testing.

That infrastructure is expensive to build, difficult to keep coherent and hard to debug.

The player expectation is much simpler: “My relationship with this character actually changed them.”

Quill knows he is safe. Astra doesn’t. CRI preserves that difference.

CRI Platform is being built to make that infrastructure reusable: a runtime plus tools to author, inspect, replay and test character state.

The next step is to validate whether other studios will adopt and pay for it, starting with teams making life and colony sims, systemic RPGs, social sandboxes and companion-led games.

How it works

Explore the storm from each character's point of view

A scene from Luminids

One thunderstorm.

Whose story are you following?

Astra's point of view

Saw Quill caught outside

What that means to her

Quill may need company after being caught outside.

Evidence
She saw Quill caught outside.
Response
She sets off to find him.

Quill's point of view

Was caught outside

What that means to him

The immediate danger has passed.

Evidence
He has found shelter.
Response
He stays put. Astra does not know yet.

Barnaby's point of view

Was somewhere else

What he knows

He does not know Quill was caught outside.

Evidence
He neither saw it nor heard about it.
Response
He carries on with his own day.

02 The architecture

The game knows what happened. CRI keeps what this character knows.

CRI is character AI middleware: memory, beliefs and relationships that stay with each character. Your choice of model gives that state a language.

Model context and character state, working together.
Character architecture For each interaction Prompt and context layer Across the character’s life CRI character state
Continuity Receives history as context for an interaction. Keeps history across interactions and saves.
Knowledge Works with the context you supply. Tracks the evidence that reaches each character.
Model role Interprets character context to generate expression. Holds the state a model gives language to.

Explore the platform

03 The proof

Remove one belief. Replay the same event. The response disappears.

Astra formed a belief through experience. We restored the same character with that belief removed and replayed the scheduled catch.

Her previous escape memories remain. Clear the belief before the shared event and the response changes.

That is the kind of causal link CRI is designed to make visible: experience changes the character, and developers can inspect the path.

See the proof Open the full run

The second catch, with and without her belief.

With her belief “I’m the fast one” She questions who she thought she was.
With that belief cleared The same catch. No belief to contradict. She is caught, but carries on.

Watch it happen in the colony

04 The product

Give characters a history. See why they changed. Test what caused it.

CRI Engine carries persistent character state. CRI Studio, Bench and Trace form the developer workflow around it: author, inspect, intervene, replay and test.

Core CRI mechanisms run inside Luminids today. The reusable platform is the productisation step that takes those mechanisms beyond our own game.

Explore CRI Platform

What CRI Platform is made of

CRI Trace The causal record connecting evidence, state, response and revision.

05 Strategic partnerships

Supported by NVIDIA and Anthropic.

06 Built in Luminids

Luminids is customer zero, not product-market fit.

It gives CRI a real production environment. The next proof is external adoption.

Luminids tests CRI against real saves, player behaviour, partial information, changing relationships and a real game’s performance budget.

It does not yet prove portable integration, willingness to pay or repeatable sales. That is the next stage of the company.

Artificial characters should have a point of view.

  • 1,000+ playtesters have played the game in testing
  • 335k Steam store visits while the game is listed as coming soon
Nick Jacob standing with folded arms in a narrow cobbled lane in Cambridge, between old stone buildings, with an iron street lamp to one side and low sun at the far end of the lane.
Nick Jacob Founder. Builds LuminidsAI and Luminids, from Cambridge.

Building a life simulation, colony simulation, systemic RPG, social sandbox or companion-led game? Talk to us about CRI.

Talk to us