Conversation analytics
Transcription can take a per-tenant vocabulary hint, and the Google Vertex AI engine can route through a gateway and authenticate with an access token.
Per-tenant transcription prompt
New feature for system administrators.
A tenant's Other tab has a Transcription settings panel with a Transcription Prompt of up to 128 characters. MiaRec Speech uses it as a hint, so a company name or product term spoken in the greeting is transcribed correctly. There is no system-wide default; a tenant without a prompt is transcribed as before.
The transcription settings of a tenant.
See Tenant transcription configuration.
Vertex AI gateway and access-token authentication
Improvement for system administrators.
The Google Vertex AI engine form gains API Server URL, API version, and an Authentication type choice between Service account JSON and Access token with a Project ID. Point the URL at a gateway such as Cloudflare AI Gateway to route, cache, or audit traffic. The Thinking Budget help text explains the 0, -1, and model-specific values, and the AI playground reports reasoning tokens for Vertex AI and OpenAI engines.
The Vertex AI engine settings with the authentication type choice.
Improvements
- Every custom field is eligible for AI insights; the separate AI assist flag on a field is retired.
- Enabling an AI task that writes to a global custom field activates that field for the tenant, unless the field was explicitly disabled for it.