> ## Documentation Index
> Fetch the complete documentation index at: https://docs.synthpop.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Voice agent capabilities
> What Synthpop's voice agent can do on a patient call — and the honest limits of each capability.
Synthpop's voice agent runs real patient phone calls at scale — **3,000+
patient calls a day and growing; the first healthcare AI agent in Google's
Gemini Enterprise Agent Gallery, running on Vertex AI.** This page is the
catalog of what it can do on a call, with the guardrail for each capability
stated alongside it.
Every capability below ships as part of the [voice
agent](/guides/engagement/voice-agent) engagement channel and is enabled per
organization during onboarding. Availability of any specific capability for
your organization is a provisioning question for your Synthpop contact, not
something you toggle yourself.
## Inbound and outbound
One Task model handles both directions of a call: the same voice agent places
outbound calls to patients and answers inbound calls from them. Whichever
direction a call runs, it yields a **transcript and summary** (and, where
provisioned, a **call recording**) returned on the Task, so the outcome reads
the same way regardless of who dialed whom.
## Speaks your patient's language
The voice agent is **language-agnostic by design**: it isn't a separately
translated flow per language, it's one flow that can run in any language.
**`en-US` and `es-MX` are live in production today**, and the production set
expands via per-org provisioning as each additional language clears quality
review — ask your Synthpop contact which languages are currently available
for your organization.
* Every call produces a native-language transcript plus a best-effort English
translation.
* FAQ answering and other tool integrations stay English regardless of the
call's spoken language.
## Answers caller questions (FAQ)
The voice agent can field a caller's free-form questions during the call. It
is **grounded in your FAQ and refuses to fabricate** an answer — but that
grounding is **self-graded, not zero-hallucination**. FAQ answering is best
suited to a small FAQ (not a large knowledge base) and is **English-only**
today.
* FAQ edits you make take effect **immediately** on the next call — there is
no separate publish or redeploy step.
## Collects a co-pay on the call
The voice agent can collect an end-of-call co-pay over the phone using
**secure DTMF** entry: **card data never reaches the model or the flow
engine**. The autopay gate **fails closed** — if it can't confirm the charge
should proceed, it doesn't proceed. This capability is **scoped to
order-creation co-pays**, not standing balances or general bill-pay. See
[Payment security](/guides/engagement/trust#payment-security) for how card
data is isolated from the rest of the call.
## Respects contact limits
Outbound dialing is bounded by guardrails on EVA's own dialing: **at most
once per day, at most three times per rolling seven days**, with an optional
local-time calling window. A call that would exceed the cap isn't placed — the
Task surfaces the reason instead — see the voice agent's how-to page for
exactly how that reason shows up.
## Speaks required disclosures verbatim
Where a disclosure is legally required, the voice agent speaks the exact
wording it was given for that language — **verbatim, not paraphrased** — so
the same disclosure reads identically on every call in that locale.
## Works with your systems
Flows can look up or act on your organization's systems through a per-org
integration, so a call can reflect live data from your side rather than only
what's on the Task. **The model never sees raw backend PHI** — see
[PHI-safe integrations](/guides/engagement/trust#phi) for how that boundary is
enforced.
## Improves over time
Call behavior can be A/B-tested and tuned without a redeploy, so how a flow
performs on real calls feeds back into how it's tuned — without a release
cycle in between.
## Every call is auditable
Every call leaves a record built for audit, not reconstructed after the
fact — see [Auditability](/guides/engagement/trust) for what that record
contains and how it's produced.
## Where to go next
The API contract for creating and reading voice Tasks.
The reviewed-flow, audit, and PHI-safety story behind these capabilities.
Where voice fits among Synthpop's engagement channels.
How engagement closes the loop after intake and coverage.