Where agents run and which agreements cover them
Agents can run inside your environment, in your cloud or on your own servers, under your controls and agreements. They connect to the systems you already use, send anything uncertain to a person, and record every action in a log your compliance team can read. This article describes where agents run, which agreements cover them, how they connect, how data is handled, and how the system is watched after go-live.
Where agents run
The system can run inside your environment, in your cloud account or on your own servers. Your security and privacy officers approve access, and your IT team approves each connection. Running there means your existing controls apply: your identity provider, your network rules, your logging and your backups.
The certifications that your cloud and model providers hold belong to them. We do not present a provider's certification as our own.
Agreements
Under HIPAA, a covered entity may share protected health information with a business associate only after getting written assurances, usually a business associate agreement (BAA), that the vendor will safeguard it (HHS, accessed October 2026). A BAA is available from Meo. A security review is available on request.
Any other data agreement your records require is discussed during the proposal. The compliance page covers consent, disclosure and vendor agreements in more detail.
Connections you keep
Your phone lines, fax, web forms, EHR, CRM, clearinghouse and payer portals stay where they are. We connect through their APIs, HL7 feeds or flat files. Where a system has no API, a supervised agent works its screens the way a staff member would, and every action it takes there is logged.
Some of these connections are about to change. Beginning 1 January 2027, Medicare Advantage plans, Medicaid and CHIP programs and plans, and plans on the federal exchanges covered by CMS-0057-F must run FHIR prior authorization APIs (CMS, 2024). We build so a payer can move from portal or fax to an API without changing the rest of the workflow.
A second route for claims
During the 2024 Change Healthcare outage, 67% of hospitals responding to an AHA survey said switching clearinghouses was difficult or very difficult (American Hospital Association, 2024). We keep a secondary clearinghouse or direct payer connection enrolled, generate claims in a format that does not depend on one vendor, and reconcile submitted, acknowledged and paid claims daily. The revenue cycle page explains why.
How data is handled
Agents get the minimum access each job needs. A coverage-check agent reads registration and insurance fields; it has no reason to open clinical notes. Recordings, transcripts and call-level logs stay in a restricted area and are processed there, inside your environment when the system runs there. When we analyze performance for our own work, we use de-identified aggregates and suppress small counts that could identify a person. None of it goes to ad platforms.
The HIPAA Security Rule may change. HHS proposed an overhaul in January 2025 and has delayed final action until at least July 2027; the current rule remains in force (Paubox, 2026). We design to the current rule and track the proposal so changes do not arrive as a surprise.
The parts of the system
Agents
Each agent has one narrow job: read a fax, check coverage, place a call, chase a payer, draft an appeal. Your teams write the rules each agent follows. We compare models on our own benchmark, including test items we keep private, and choose per task (how we test models).
Human review
Anything clinical, uncertain or above a threshold you set goes to a person, who approves or changes it. Clinicians, coders and managers make those calls. The human review page shows the four decisions that always stay with your people.
Audit log
Each action is recorded with its time, its input, the evidence it used and its outcome. The log exports to your own storage, and your compliance team can review any action.
Monitoring
Scheduled routines watch what was built and alert a named owner on your side when something drifts, for example a payer portal that changed its layout or a spike in calls passed to staff. Routines typically run in shadow for 14 days before they go live. Alerts go to email, chat or your service desk.
If you want to change course
Every phase ends with your sign-off, and any line can switch back to today's process at any time. When the system runs in your environment, your data stays in systems you control. Ask for the ownership and exit terms in your proposal before you sign.