> ## Documentation Index
> Fetch the complete documentation index at: https://docs.meshpilot.app/llms.txt
> Use this file to discover all available pages before exploring further.
# Voice Agent
> Inbound and outbound phone calls — COD confirmation, feedback, lead-qual.
The Voice agent runs your phone-call layer through LiveKit and your telephony provider. The most common use case today is **COD (cash-on-delivery) confirmation** for D2C brands shipping to markets where COD is the dominant payment method.
## What it does
The Voice agent runs scheduled outbound batches (confirm today's COD orders, follow up with abandoned-cart customers, run feedback surveys on recent deliveries) and handles inbound calls where configured. Each call uses a brand-voice TTS layer plus a tool-using LLM that can read context (the order details, the customer's history, the script) and respond naturally.
## What it proposes
After each call, the agent summarises the transcript and writes a proposal:
* **Confirm order** — clean confirmation, low-fraud signal, customer confirmed delivery slot.
* **Reschedule delivery** — customer requested a different window.
* **Cancel order** — customer declined or could not be reached after retries.
* **Escalate to human** — fraud signal, complaint, or ambiguity the agent doesn't want to commit on.
## What you approve
The Voice agent never silently mutates an order. Even on a clean confirmation, the proposal is queued — though once trust is built, this is one of the first action types where operators dial up autonomy (clean COD confirmations under daily cap, no escalation flags).
## What the Brain learns
Pickup-rate patterns by time-of-day and weekday, language preferences by region, common fraud signals, script lines that won customer trust vs caused friction, and operator overrides on close calls.
Voice minutes are metered per plan. See **Pricing → Usage limits** for current pool sizes.
See your call log.