What this is

Everything the agent knows before it dials

For every guarantor who owes money, we pull their record from Denticon, decode and redact it server-side, and hand the voice agent one compact collection_context object. The agent uses it to explain the debt and answer common objections — without ever touching a raw EHR record.

0
Denticon endpoints available
0
Core endpoints for a minimum agent
0
Fields currently marked to keep
0/patient
Marginal API calls in batch mode

How a record becomes a call

One server-side step does the pulling, decoding and redacting — the agent only ever sees the green box.

01 · SOURCE
Denticon EHR

Balances, ledgers, claims, coverage, notes across 22 offices.

02 · SOURCE
Batch fetch

Office-wide pull, indexed locally by patientId / responsiblePartyId.

03 · SOURCE
Decode + derive

Resolve ledgerType, build last payment / charges / claim status.

04 · SOURCE
Redact

Mask subscriber IDs & checks, summarize notes, attach guardrails.

05 · SOURCE
Voice agent

Smartflow injects it as call variables; agent speaks within guardrails.

What this dashboard answers

The four questions behind the build.

1Which endpoints to hit per patient, and how many calls it costs.
2Every variable we can pull, what it means, and how it ties to the debt.
3How to pass those variables into the agent prompt.
4A concrete sample payload for one patient.

Where to look

Navigate using the sidebar.

BBusiness — Variables explained, the sample payload, and objection use-cases. Use the keep/drop toggles to lock in what actually ships.
TTechnical — The 12 endpoints, call-count math, auth, and exactly what JSON gets injected into the LLM.

⚑ Decision needed: 60+ raw fields are available, but the agent should not get all of them.