Skip to content

Feature · PII redaction

PII redaction for voice agent sessions and transcripts.

PyAI screens each voice agent reply before it is spoken, in an app, on a website or on a phone line, and Trace checks finished transcripts for personal data.

Get a sandbox key$50 credit. No credit card required.
Guardrail before every spoken replyTrace scan once the session endsAudit hash per scored sessionTrace $0.12/call

What PyAI gives you for PII redaction

  • PII redaction also reaches batch work: Hear transcription jobs with Trace switched on get the same PII scan and keep a redaction verdict beside the text.
  • Masking works on text only, so an agent that must retain no audio should have recording turned off, since PyAI does not document audio redaction.
  • Trace opens in warn-only mode at $0.12 a call, so findings appear before any rule blocks a reply.
  • A card number the user reads aloud is caught by Trace after the session rather than masked mid-stream, because the live guardrail screens agent speech only.

What happens when a reply or transcript contains PII?

Select one of three moments and run it. Each line is invented for illustration and comes from no real user.

Pick a moment, then press play

UserAI voice agentGuardrail and Trace
Ready00:00

How does PyAI PII redaction work, live and after the session?

Personal data is checked at two points: the guardrail screens what the agent is about to say, and Trace screens the finished record.

Reply drafted

During an Omni session the reasoning model writes a reply from persona, knowledge facts, the caller card and conversation history.

Output screened

Ahead of speech synthesis the guardrail checks that output and can mask PII or stop the reply, alongside persona and tool rules.

Record stored

When the session ends PyAI writes one durable record plus a stereo recording when allowed, and agents with recording off have the audio dropped.

Trace scores it

Trace grades the conversation on TCPA, HIPAA and PII. Reviewers open it to see violations, passed checks, redactions and the audit hash.

An active rule matches

A review opens, the scorecard-completed webhook fires and alerts go to Slack or your webhook.

Still in warn-only mode

The finding is logged with a warning and nothing is blocked, leaving room to preview rules on recent sessions first.

Which teams need a PII redaction tool for voice agents?

A PII redaction tool matters most when transcripts leave the voice stack. That includes QA tools, CRMs, reports and AI model training.

Healthcare intake tablet

Patients naming conditions

Intake sessions on a clinic tablet surface symptoms and medication names. Trace checks each session against HIPAA rules and logs redactions with an audit hash for compliance staff.

In-app billing assistant

Card numbers spoken aloud

An agent can confirm the card ending 4279 while the guardrail keeps the full number from ever being read back.

Outbound phone sales

Consent checks

Trace looks for TCPA findings on every scored call, so a missing consent line lands in the review inbox rather than in an audit.

Voice agent QA

Transcripts with many readers

Trainers, reviewers and CRM syncs all touch transcripts, and scanning first keeps personal data out of places it should never reach.

AI product teams

Data leaking into model prompts

A frequent question is whether masking at the speech-to-text layer is enough when PII can also slip into language model prompts and logs. On Omni the guardrail screens the model's own output too.

PII redaction questions people ask

What are 5 examples of PII?

Five common examples of PII are a full name, a phone number, an email address, a payment card number and a government ID such as a Social Security number. Voice conversations also surface health details and home addresses, which a redaction step should mask before any transcript is shared.

Does PyAI redact PII during a live conversation?

Yes, for agent replies. In every Omni session a guardrail checks each reply before speech synthesis and can mask PII or block the reply. User speech is not masked mid session, and Trace scans the complete transcript afterwards and records findings, redactions and an audit hash.

Does PyAI redact PII from session recordings?

Not in the audio, as documented today. Trace records PII findings and redactions on transcripts and session records, but PyAI docs do not describe muting or bleeping the recording. Teams that must keep no audio can turn recording off on the agent, and PyAI then drops the file.

How much does PyAI redaction cost?

The live guardrail comes with Omni at $0.05/min plus $0.01 telephony, with no platform fee. Post-call PII scanning is part of Trace, an add-on at $0.12/call that also scores TCPA and HIPAA. New accounts begin with $50 of credit and need no credit card.

PyAI is voice infrastructure. It hands you the audio and transcript, and does not replace your own models or review tools.

Keep card numbers off the line and out of every transcript.

Start building with $50 creditSandbox keys never bill, and you can start without a credit card.

One stack: Omni, Hear, Speak, Recap and Trace.