Line health & reliability
A phone line is infrastructure. We built it like infrastructure.
CallsAround verifies every line is answerable by holding real conversations with it around the clock, and backstops the entire AI stack with plain carrier call forwarding at the phone network — a failover that works precisely when our servers don't. This page explains the architecture, layer by layer.
How does CallsAround verify a line can actually answer?
Most monitoring asks "is the server up?" — a question customers never ask. The question that matters is: if a customer called this exact line right now, would the agent handle it? So that's the question we test. A conversational canary probes every line around the clock by holding a scripted conversation with the line's own production agent — through the same telephony path a real call takes — asking to reach a human and verifying the agent responds correctly, tool calls included.
Because the probe exercises the full stack — telephony provider, AI model, our dispatch engine, and the line's specific configuration — it catches failure modes a health-check endpoint structurally cannot: a degraded model returning empty turns, a misconfigured agent, an upstream voice provider having a bad night. One failed probe is a data point; consecutive failures are an incident, and the system acts on its own.
The layers, and what happens when each one fails
A call passes through four layers. The design rule: every layer's failure is caught by a mechanism that lives below it.
| Layer | Failure mode | What catches it |
|---|---|---|
| AI voice model | Empty turns, degraded responses, provider outage | Canary conversations fail → consecutive-failure threshold → carrier failover |
| CallsAround app & dispatch engine | Server outage, bad deploy, database incident | Canary can't complete → carrier failover engages at the network, which does not depend on our servers |
| Telephony provider | Call routing or media degradation | Canary rides the same path and fails the same way → failover + alerting |
| The carrier itself | The phone network — the layer everything trusts | This is where the backstop lives: plain call forwarding to the business's backup phone, the same mechanism every phone number on earth relies on |
Why is carrier-level failover different from an API-only architecture?
Most AI answering products are built entirely above the telephony API: the number lives at a VoIP provider, calls are handed to the vendor's application, and the application drives the AI. It's a fine architecture — until the application layer is the thing that fails. Then the call connects to silence, an error tone, or an infinite ring, because every mechanism that could have saved it lived in the layer that just went down. The vendor's own outage disables the vendor's own recovery.
CallsAround puts the backstop one layer lower. When the canary detects repeated failures, the business number is switched to plain carrier call forwarding at the phone network — below our servers, below the AI, below the APIs. Callers ring the business's backup phone the way phone calls have been forwarded for decades. The failover's job is to work when nothing above it does, and it's the only place in the stack where that property can actually hold.
Recovery is automatic and symmetric: probes keep running against the line, and when they pass again, forwarding switches off and the AI takes the line back — with the owner alerted at both transitions and the whole episode on the public record.
Failing loudly: alerting and the public record
Reliability isn't just staying up — it's never failing silently. Every transition is announced and journaled:
- Owner alerts by email and SMS when failover engages and when the line restores.
- A public status page at callsaround.com/status driven by the canary itself — live health and incident history, not a static badge.
- An escalation journal on every emergency call: who was rung, when, what answered, who accepted — timestamps, not memory.
- Uncovered-call alerts: if an emergency exhausts the on-call chain, the whole team is told. An unanswered emergency is an incident, never a shrug.
Canary monitoring, the status page, and alerting cover every line on every plan; automatic carrier-level failover is included with Dispatch+.
Reliability questions, answered straight
Verify us before you trust us with the line.
Watch the status page, read how dispatch works, then put us on your overflow calls first — instant self-serve setup, month-to-month pricing.