Prompt and Keyterm Support Now Available for AssemblyAI on Telnyx Voice AI Assistants
20, Oct 2026
Voice AI Assistants that use AssemblyAI Universal-3.5 Pro for speech-to-text can now pass a context prompt and a list of keyterms to the model. Telnyx also sends the agent's last reply to AssemblyAI after each turn, so the model hears every caller response with the question it answers.
What's new
Prompt: Describe the call in up to 1,750 characters, such as the domain, names, and what callers usually talk about. AssemblyAI uses it to recognize the vocabulary that fits the conversation.
Keyterm Boost: Add up to 100 comma-separated terms, each up to 50 characters, to improve recognition of product names, drug names, and other domain vocabulary.
Agent context: After each reply, the assistant passes its last spoken reply to AssemblyAI automatically. No setting is required.
Why it matters
Domain vocabulary is where speech-to-text errors cost the most. A misheard product name or medication sends the whole turn down the wrong path.
Keyterms and the prompt are optional with Universal-3.5 Pro, so an existing assistant keeps working without changes.
Prompt, keyterms, and agent context travel through the same Telnyx pipeline as the rest of the call, so there is no separate STT vendor integration to build or maintain.
Example use cases
Healthcare scheduling agents that need drug names and clinic locations transcribed correctly.
Support agents for a product catalog with part numbers and brand names callers say out loud.
Financial services agents that handle account types, fund names, and ticker symbols.
Getting started
In Mission Control, go to AI, then Assistants, and select your assistant.
Open the Voice tab and select assemblyai/universal-3-5-pro as the transcription model.
In Prompt, describe the content of your calls. Describe the content only, not instructions for the model.
In Keyterm Boost, enter your terms separated by commas.
Save your assistant.
Requests over the limits return a 400 error from the API.