Resolved on March 4, 2026, affecting section-306293 (OpenAI Chat Completion) and section-306294 (Claude Chat Completion).
After running experiments with Enterprise Managed users and white label users (with their permission from support inquiries), we noticed that user BYOK on voice is significant increasing latency, creating scenarios of no response, and potential hallucinations. This is because user keys may not be "seasoned" or high enough tier to support high frequency of API calls (aka voice session) which could cause 429 errors and 500 errors. As a temporary solution, we are using our key as a fallback until we can engineer a tiered system for checking API keys. The idea we have engineered is to check keys for their tier / limits to give them eligibility for voice use. More to come.
Affected components
Update timeline
- resolved Mar 04, 2026, 07:36 PM UTC
After running experiments with Enterprise Managed users and white label users (with their permission from support inquiries), we noticed that user BYOK on voice is significant increasing latency, creating scenarios of no response, and potential hallucinations. This is because user keys may not be "seasoned" or high enough tier to support high frequency of API calls (aka voice session) which could cause 429 errors and 500 errors. As a temporary solution, we are using our key as a fallback until we can engineer a tiered system for checking API keys. The idea we have engineered is to check keys for their tier / limits to give them eligibility for voice use. More to come.
Your product runs on more than Assistable
Add the vendors you depend on and get one message when any of them reports trouble. Free for 3 vendors, no card.
Amazon Web Services