Independent AI lab
Building agents, tools, and experiments that hold up.
We build interfaces, data tools, coding systems, and weird intelligent products until they either become useful or get replaced by something sharper.
Agents in the wild
We build assistants, workflows, and interfaces that have to survive actual use, not just a launch tweet.
Useful weirdness
A lot of the work starts as experiments. The good ones turn into products. The strange ones keep cooking.
Rogue, not random
Fast iteration is the culture, but the bar is still usefulness, clarity, and systems that hold up under pressure.
Shipping now
Major focus right now is /Slash and Apex.
ACRA, TagPilot, Pulse, and ChatPilot continue shipping in parallel.
Current research
ACRA: Auditable LLM Agents for Evidence-Grounded Clinical Decision Support
Building agents that surface traceable evidence, transparent reasoning steps, and clinician-legible outputs.
In the lab
BrainCrate, Aureus, ACRA, and whatever interesting model behavior survives the week.
Experimental tracks stay small, fast, and disposable until they prove useful.
/Slash
Apex
ACRA
Evidence-grounded agents
Clinical decision support