Real-world evaluation
for AI characters.
Powered by experiences where people get to know them and shape stories.
What we're building
Typicity builds original AI character experiences alongside commercial evaluation tools for AI teams. Everything we do comes back to one question. When a model updates underneath, does a character still feel like itself?
Anyone who has spent real time with an AI character knows the frustration of watching it suddenly lose its voice after a system update. Standard automated tests rarely catch that shift, because they only check for predefined bugs. Recognizing whether a character is genuinely growing or drifting into a stranger takes real familiarity over time.
Through our interactive stories, players get to know our original characters in depth. Alongside the story, players can optionally join structured evaluation sessions to share their judgment on how a character holds up across model updates. By pairing this human perspective with our automated long-horizon checks, we help AI teams see where a model change crosses the line from natural growth into replacement before they ship. Private conversations stay strictly private and never enter any commercial evaluation.
Current stage. Our underlying memory and regression engine is up and running, and our first playable story chapter is in active development. The concept film below shares the world we're creating for players, while the interactive example illustrates our evaluation approach for AI teams.
Characters worth
knowing over time.
Each character arrives with a life already in motion. The relationship can change them, but it does not erase who they are.
Would they still make
the same choice?
An update can make the conversation smoother and still make the character feel wrong. You notice the difference when you know them.
A say in what happens next.
At the beginning of a difficult journey, the character makes a promise.
“If we change the plan, we decide together.”
We’re building evaluations that combine automated checks with reviews from people who know the character. Teams could follow how behavior changes across models and over time.
Private relationships
stay private.
Private conversations and personal relationship history stay outside commercial evaluation.
Contributing is always your choice. Only separately opted-in, structured story judgments enter the evaluation path. Saying no never changes how the story treats you.
Building AI with
a sense of character?
We’d like to hear what consistency means for your team.
Get in touch