SOFTPHONE PRO · INTERNAL BULLETINIssue 04 · November 2026
Softphone Pro
Softphone Pro
Business call management · Bulletin
Published by the editorial desk
Updated 2026-11-20
Product & Technology · Issue 04

From cloud PBX to AI Agent Studio: the piece Cloudya is adding

Connecting telephony to business used to mean APIs and middleware. Now there is another route: letting an AI agent take part in the call flow directly.

2026-10-30Engineering DeskApprox. 6 min read

Cloudya is NFON's cloud communications platform, covering cloud PBX, SIP trunking, endpoints, contact centre, integration and automation. In the past two years the portfolio gained a new member: AI Agent Studio.

The problem it targets

Enterprises have always wanted to automate phone flows, and historically had two options: configure an IVR inside the PBX, which is limited to key presses, or build an integration, which is expensive and demands telephony knowledge.

AI Agent Studio offers a third route: define an agent visually. It understands natural language, looks things up according to rules, answers, and hands over to a human when necessary.

Three typical uses

  • Reception triage: ask the reason for the call, decide whether it is a visitor, supplier or customer, and route accordingly.
  • First-line support: answer standard questions such as order status or rescheduling, escalating anything complex.
  • Out-of-hours cover: answer in the evening and at weekends, record the request, and generate a task list for the morning.

A voice bot succeeds or fails on whether it knows what it does not know.

Real engineering constraints

Voice interaction is much harder than text: it must handle interruptions, accents and background noise, and it must control latency — a silence longer than a second makes users think the line dropped. That means model calls cannot be slow and local fallback phrasing is mandatory.

The other constraint is explainability. Enterprise customers need to know why the bot answered as it did, and to trace it afterwards. That requires logging across the whole chain: recognition, understanding, retrieval, generation.

Relationship to the traditional IVR

Not replacement but layering. High-frequency fixed paths still suit keypad IVR: fast, deterministic, no errors. Natural language suits the long tail of people who cannot articulate what they need. The mature pattern mixes both — offer keypad options while allowing speech at any point.

Questions to ask when buying

Where does the model run? Does call data leave the premises? Can it use our own knowledge base? Can escalation rules be customised? How far do the logs let us reconstruct a failure? Those answers matter far more than a demo.

CloudyaAI AgentEcosystem