Browser call (WebRTC)
Call state
Live transcript
PII redaction
Controls running during the call
This panel is an original mockup — not real call data
Safety processors loading
Pipeline load
In-call checks
Failure modes handled
Safety processors loading
Pipeline load
In-call checks
Failure modes handled
Browser call (WebRTC)
Call state
Live transcript
PII redaction
Controls running during the call
This panel is an original mockup — not real call data
A voice call is unforgiving, so the failure modes are handled ahead of time rather than hoped away. Before the call starts, while it runs, and after it ends — each stretch carries its own controls, and security here is scoped to within the governed agent workflow.
One path, controls on all three stretches
What is open and what is locked is decided before anyone dials.
The safety processors run together in a single pass over the stream.
The record and its follow-up hand off to the console, and production actions run only after a person confirms them.
PII redaction, output guardrails, and conversational risk signals run as processors inside the real-time pipeline. They are enabled per deployment, and redaction is not a step bolted on afterwards.
On call
Running in one pass
Call session pipeline
Degrade behaviour
When something is uncertain the pipeline steps a capability down instead of dropping the call.
The guard wrapper refuses to deploy if a live function would vanish, non-interactive runs fail closed, and a typed override is logged to evidence. Every unverified state resolves to “no.”
The release pathFail-closed
| Step | State |
|---|---|
| Agents propose | Pull request |
| Checks run | Passed |
| Guard evaluates | Held |
| A person approves | Waiting |
Transcripts and recordings land in the console after the call. Retention, storage location, and access are scoped with you per deployment rather than assumed.
Call record
Recording
Transcript
Follow-up
Runs after a person confirms
Browser and mobile voice are the transports live today. The telephony architecture is built but key-gated and is not running as production phone calls — we would rather say that plainly than imply you can dial in tomorrow.
Transports
| Path | State |
|---|---|
| Browser (WebRTC) | Live |
| Mobile | Live |
| Telephony (PSTN) | Key-gated |
| Console API | Per deployment |
Telephony is integrated during a deployment rather than switched on out of the box.
30+
custom voice-pipeline processors
100+
feature flags
450+
operator-console components
34
care and consumer backend domain modules
Every figure was counted in the repository — audit, August 2026.
Vendor failover, silence watchdogs, and feature flags hold the call instead of dropping it. The last step into production is always a person.
Figures below are from the August 2026 repository audit
Review the controls one at a time. In-handler authentication, the fail-closed deploy guard, a machine-checked topology check, and a self-auditing CI — we're happy to walk your security team through them directly.
Follow a call from connect to close. Transport connect, the safety processors, the degrade behaviour, the call record, and the human confirmation — the real sequence, not a simplification.
IntuneLabs
Shortcuts
Areas
Safety
Release evaluation requests
Release evaluation requests return a result for each change. An unverified state is treated as not deployable.
All requests
Sample
Held
Sample
Reference
Certification statusWe hold no SOC 2, HIPAA, or GDPR certification.
When something is uncertain the pipeline steps that capability down. Risky paths default off, and the system holds the call instead of ending it.
Sticky failover between text-to-speech providers means one vendor degrading mid-call doesn't end it.
We orchestrate best-of-breed vendor models for speech, language, and synthesis. What we build is the Korean tuning, the orchestration, and the reliability layer around them.
IntuneLabs
Certification status, stated first: We do not currently hold SOC 2, HIPAA, or GDPR certifications. What we offer today is inspectable engineering controls, and we're happy to walk your security team through them directly.
See deployment scenarios
Not yet. We do not currently hold SOC 2, HIPAA, or GDPR certifications.
What we offer today is inspectable engineering controls — a fail-closed deploy guard, in-handler authentication, and CI checks on every change — and we're happy to walk your security team through them directly.
Unverified states resolve to “no.” Non-interactive deploy runs fail closed, the deployability check treats unknown as not-deployable, and dangerous actions are permission-blocked rather than merely discouraged.
No. Deploy, merge, and auth actions are permission-blocked for coding agents, and the deploy guard refuses to self-approve in non-interactive runs.
Agents open pull requests; a human is the single gate to production.
PII redaction runs inside the pipeline as one of the processors. Retention, storage location, and access for a given deployment are scoped with you during the engagement rather than assumed.
Not out of the box. The pipeline ships a telephony-ready architecture, but the transport modules are key-gated and are not running as production phone calls.
Telephony is integrated per deployment. Browser and mobile voice are the transports live today.
We don't train our own speech or language models — we orchestrate vendor models and build the Korean tuning, orchestration, and reliability layer around them.
No. A voice agent is not an emergency service and must not be relied on as one. In an emergency in Korea, call 119 for fire or medical help and 112 for police.
Autonomy up to the pull-request boundary, fail-closed everywhere past it. Talk to us about what that means for your deployment.
Scoped per deployment
What is switched on, what stays locked, and how records are handled are decided with you, deployment by deployment.
See pricingIntuneLabs does not currently hold SOC 2, HIPAA, or GDPR certifications, and nothing on this page claims a certification or regulatory compliance. Security here is scoped to within the governed agent workflow, and transport-level authentication is assumed absent by design.
Every panel on this page is an original mockup. None of them shows a real call, a real person, or real customer data, and we do not publish latency benchmarks — per-call timings are instrumented for operations. Figures are limited to what the August 2026 repository audit counted.