Patient access, without the data leaving your network In development
Scheduling, intake, benefits verification and pre-visit instructions, running entirely inside your own infrastructure. We are taking design-partner conversations now, not production deployments.
What it does
Scheduling and rescheduling
Slot lookup, booking and change requests, written back to your system with a structured record.
Intake questions
Structured collection of pre-visit information, in the patient's language, with the script correct for downstream systems.
Benefits verification
Grounded lookup against your own coverage data, retrieved rather than recalled.
Structured handoff
A clean summary for staff, with everything the patient was told, logged and reproducible.
Patient audio is the strictest version of the problem we build for
Health information on a phone call is the case where sending data to a third-party model is hardest to justify and hardest to get approved. There is no vendor agreement that makes it as simple as never sending it at all.
Self-hosting is not a preference here. For many providers it is the only architecture that clears review, which is why this is the clearest expression of what we build — and why we want to get it right before we sell it.
What we are honest about: we have not shipped this. If you are a hospital group or provider network with call volume and an infrastructure team, a design-partner conversation is genuinely useful to both of us. If you need something in production next quarter, we are not it.
Who this fits
Good fit for a design partnership: hospital groups, diagnostic chains and provider networks with high call volume, existing recordings, an infrastructure team, and a compliance function that has blocked hosted AI.
Common questions
Is this available now?
No. Healthcare patient access is in development and we are only taking design-partner conversations. We would rather say that than sell you a roadmap.
How is patient data handled during fine-tuning?
In an environment you control, or one agreed in writing. We do not train shared models on customer conversations, and your model is not used to improve anyone else's.