Date: 2026-08-26
Impact: 15:15–17:58 UTC
Description of Issue
Customers experienced elevated latency on Text to Speech time-to-first-audio, Speech to Text session start, and Agents Platform conversation setup.
Root Cause
We received a burst of concurrent requests to a configuration read endpoint on the Agents Platform. The internal validation step behind that endpoint scales with the number of related objects in an account, and this account had accumulated an unusually large number of them. Each request therefore expanded into a very large amount of internal work against a shared caching layer that ElevenLabs services rely on.
That caching layer became saturated. Because it is shared across products, every service waiting on it slowed down.
Requests continued to succeed throughout, this was a latency degradation, not an outage. No requests were rejected as a result of this incident, no data was lost, and access controls behaved correctly at all times.
Timeline (UTC)
15:15 First burst of concurrent requests begins; internal lookup latency starts rising
15:38 Incident declared; investigation begins
15:50 Degradation localized to a shared internal caching layer
17:50 Service recovers; latency returns to baseline
17:58 Full recovery confirmed across all affected products
Preventative Actions & Learnings
Removed the repeated internal lookups and narrowed the associated database reads
Batched the remaining reads and exits the check early once access is established
We apologize for the disruption this caused to your workstreams and are committed to preventing similar issues moving forward.
