Question 1

We already generate the note or data. What work is still left inside the EHR?

Accepted Answer

Generating the note isn't filing it. The scribe's output still has to land in the right patient, encounter, and document type inside the EHR, and extraction has the same gap in reverse. Agents write the note into the chart the way a person would and pull authorized fields back out, so neither direction stalls on a system with no usable API. The AI scribe write-back deployment is the production version of this.

Question 2

Can this run through browser, desktop, or Citrix environments without a new interface?

Accepted Answer

Yes. Web EHRs run as browser agents; legacy and Citrix-hosted desktop EMRs run as computer-use agents driving the same click-by-click interface a person uses. Asteroid deployed the first computer-use agents against Citrix-hosted EMRs in healthcare. No new interface, no vendor integration program, no waiting on an API that isn't coming. One production deployment extracts 40,000 patient records a month across 11 different EHR systems this way. More in the desktop automation breakdown.

Question 3

What happens when the patient cannot be matched or the target field is unavailable?

Accepted Answer

The write doesn't happen. A patient, encounter, or destination that can't be matched above your confidence threshold stops the run and routes to a person with the candidates found. A note filed to the wrong chart is the error this workflow exists to prevent, so ambiguity always loses to escalation. A field locked by permissions or absent from the form comes back as an explicit exception naming the field, never a silent skip.

Question 4

What happens when the EHR changes its layout, permissions, or write workflow?

Accepted Answer

Layout changes are absorbed by reading the screens live (including inside Citrix sessions) rather than replaying coordinates. A permissions change that blocks a write, or an upgrade that restructures the documentation flow, stops runs with the block identified; pending writes queue rather than disappear. Asteroid maintains the workflow against the new version: an EHR upgrade is our maintenance event, not your team's broken script.

Question 5

How are PHI, credentials, write permissions, approvals, and audit history handled?

Accepted Answer

Extraction returns only the field set you authorize; writes run under EHR user accounts you control, scoped to least privilege and revocable at any time. Any write class you designate (new documents, order-adjacent fields) can require human approval before saving. Every read and write lands in execution history with timestamps, on HIPAA-compliant, SOC 2 Type II infrastructure under a BAA. The full access-control posture is on the security page.

Question 6

What if an EHR or module we need is not in the library?

Accepted Answer

If a person can operate it through a screen, it's scopeable: map the workflow in the builder against your actual chart-prep or extraction steps, or have Asteroid build and operate it managed. The 11-EHR production deployment above grew exactly this way: system by system, each addition reusing the same matching, approval, and evidence machinery already proven on the others.

Read and write the chart, end to end.

Close the loop inside the EHR.

Frequently asked questions