Data residency
Which parts of a call run in Australia, where call data is stored, and how a workspace's residency setting governs fallback. The sub-processor list is available on request.
Speech recognition, the default voice and the default AI model run in Australia. Recordings and transcripts are stored in Sydney. Each workspace's data residency setting decides whether a call may fall back to an AI model or voice outside Australia when the Australian one fails.
What runs in Australia
| Part of the call | Region |
|---|---|
| Voice runtime compute | Sydney |
| Web application and API | Sydney |
| Speech recognition, including its fallback | Australia |
| The voice, default | Australia |
| The AI model, default | Sydney and Melbourne |
| Content moderation | Sydney |
| Database and file storage | Sydney |
| Database backups | Sydney |
The residency setting
Each workspace has a data residency setting with two values. Every workspace is set to Australia unless a workspace owner or admin allows overseas fallback.
- Australia. When the AI model or the voice fails, the call has no overseas backup, and neither does the analysis after the call.
- Overseas fallback allowed. When the Australian AI model fails, the call can fall back to an overseas provider of the same model. To keep that provider ready, the agent's instructions are sent to it at the start of each call. An agent with an overseas fallback voice can use it.
Only a workspace owner or admin can change the setting, and each change is written to the workspace audit log. Either way, an agent you switch to an overseas model or voice by name uses it. Failover covers what happens when a provider fails.
Language models
The default AI model runs in Australia, and so do several of the other available models. The rest are served from outside Australia, and a call reaches one only if you choose it for an agent, or as a fallback where the workspace allows overseas fallback. The model picker labels each model's residency, and Available models covers the tiers.
A model's region is not the whole picture for a voice call: the voice is covered separately below.
The voice
The default voice runs in Australia. If you switch an agent to an overseas voice, the agent's reply text is synthesised outside Australia and the audio comes back. This matters most when the agent reads personal details aloud, because those details are the reply text. See Voice to change an agent's voice.
Telemetry
Error tracking, product analytics, model tracing and log aggregation receive metadata only by default.
Sent by default: identifiers, counts, durations, model names,
error types, stack frames
Not sent by default: transcripts, prompts, tool arguments,
personal informationTelemetry content levels
| Level | Behaviour |
|---|---|
| Off (default) | Content attributes removed before export. |
| Redacted | Content passes through PII redaction before export. |
| Full | Content exported verbatim. |
Set per workspace with an optional per-agent override. The level applies per call, so one workspace's setting cannot affect another's exports.
Recording app users' IP addresses in product analytics, and session replay, are both off by default.
What is stored, and where
| Data | Location | Retention |
|---|---|---|
| Transcripts and call metadata | Sydney | Kept until the workspace or the agent sets a retention period |
| Call recordings | Sydney, private storage | Kept with the call's transcript, until a retention period is set |
| Caller profiles | Sydney | Until deleted, or until a retention period forgets the caller |
| Knowledge base content and embeddings | Sydney | Until the source is removed |
| Agent settings | Sydney | Until deleted |
| Secret variables | Sydney, encrypted with a separate key | Until deleted |
On a phone call, the phone-line provider makes the recording and we copy it to our storage in Sydney. Once our copy is verified, we delete the provider's copy and check that none is left.
See Privacy for retention and deletion details.
Sub-processors
Our sub-processor list is available on request. Email privacy@verticalai.com.au with the subject line "Sub-processor list". It names each provider, what it does, the data it handles and its region. Customers get 30 days' notice by email before a sub-processor is added.
Verifying this for your workspace
Your telemetry content level is shown in workspace settings, under Privacy and retention, as Call content in our support tools. For anything else, contact support. We can provide written confirmation of data processing locations for a security review.