Research & development

Questions we are actively working on.

Research is where the honest labels matter most. Each of these is a live question, not a product page with a paper taped to it.

Harness design
In development
What is the smallest scaffold that makes a model dependable on real work?
Agent reliability
Research
How do you measure an agent’s behaviour when nobody is watching it?
Orchestration at the seams
Research
Where do multi-agent systems actually fail — and can it be designed out?
Constrainable hardware
In development
What does a wearable need to prove before a person should trust it with their identity?
Edge intelligence
Research
How much useful intelligence fits in a device that must run on a coin cell?
Compute economics
Research
What does this cost per useful answer, and can it be honest about that?

Method

How we work, in four sentences.

Honest by default
We would rather write “in development” under a heading than imply a product we have not shipped. Every capability on this site carries its real status.
Systems, not demos
A demo answers a question once. A system answers it on a Tuesday, with a real customer, when something upstream has changed.
Humans hold the keys
Approval, escalation, and the ability to stop are features, not friction. Autonomy is earned in small, reversible steps.
Serious about the boring parts
Cost, failure modes, recovery, and observability. The parts nobody films are the parts that decide whether it survives contact with reality.

Research collaborations, prototypes, and the occasional beautiful dead end.