Feature Type
Nice to have
Feature Description
Hey I’m building an voice agent and running into a bit of a privacy hurdle.
Right now, it looks like livekit-agents pipes the transcribed text straight from the STT engine directly to the LLM. It would be super helpful to have a native middleware hook or interceptor right in the middle of that pipeline.
The Use Case:
For privacy and compliance, we need to scrub sensitive info (like credit card numbers, SSNs, Aadhaar, etc.) before the text leaves our servers and hits external APIs like OpenAI or Anthropic.
If we had a pre-LLM hook, developers could easily plug in a PII redaction tool (like Microsoft Presidio or custom regex) to mask the text on the fly.
Basically, the flow would look like this:
User Speaks ➔ STT ➔ [ Custom Middleware / PII Scrubber ] ➔ LLM
Feature Type
Nice to have
Feature Description
Hey I’m building an voice agent and running into a bit of a privacy hurdle.
Right now, it looks like livekit-agents pipes the transcribed text straight from the STT engine directly to the LLM. It would be super helpful to have a native middleware hook or interceptor right in the middle of that pipeline.
The Use Case:
For privacy and compliance, we need to scrub sensitive info (like credit card numbers, SSNs, Aadhaar, etc.) before the text leaves our servers and hits external APIs like OpenAI or Anthropic.
If we had a pre-LLM hook, developers could easily plug in a PII redaction tool (like Microsoft Presidio or custom regex) to mask the text on the fly.
Basically, the flow would look like this:
User Speaks ➔ STT ➔ [ Custom Middleware / PII Scrubber ] ➔ LLM