Tavus Outage History

An infrastructure issue prevented some conversations from starting on Tavus's Conversational Video Interface. The problem affected users' ability to initiate new video conversations for 27 minutes. The issue was resolved after implementing a fix and monitoring the results.

Tavus experienced elevated error rates across multiple services including perception, knowledge base, Gemini conversation LLM, objectives, and guardrails due to their cloud provider throttling model inference. The Conversational Video Interface was affected for 5.3 hours while the team worked with their cloud provider to restore deployments and resolve the throttling issue. All services were fully restored and the incident was resolved after deploying fixes and restoring all deployments.

Tavus experienced increased error rates with their Raven-0 and Raven-1 perception models, affecting the Conversational Video Interface for 5.2 hours. The issue was identified and a fix was deployed, with the team monitoring system stability before confirming full resolution.

Tavus experienced intermittent call connection issues for 3.8 hours due to upstream outages at Daily and Cloudflare, affecting the API and Conversational Video Interface. The service disruption prevented users from reliably establishing video calls through the platform. The issue was resolved after Cloudflare implemented a fix, allowing downstream services to return to normal operation.

Tavus experienced an issue where replicas were taking longer than normal to join conversations, affecting the Conversational Video Interface. The problem lasted 38 minutes before being resolved. The service team investigated and successfully fixed the issue.

Tavus experienced elevated API errors from their webRTC provider Daily, causing issues with the Conversational Video Interface and preventing users from joining conversations properly. The incident lasted 18 minutes and was resolved after Daily's team released a fix. Traffic returned to normal levels and the service was fully restored.

Tavus experienced a 23-minute incident where conversation requests succeeded but replicas failed to properly join due to GPU-level issues caused by a sub-provider's AWS instance outage. The Conversational Video Interface was affected, with some conversations that were routed to the impacted provider unable to join properly. Traffic was automatically rerouted to restore normal functionality within SLA parameters.

Tavus experienced slow conversation boot times due to queuing issues with an underlying GPU provider, affecting the Conversational Video Interface for some users. The team resolved the issue by shifting traffic away from the problematic provider. The incident lasted 33 minutes and was fully resolved.

Tavus experienced a major incident where some conversations were failing to initiate, affecting the Conversational Video Interface. The service team identified the root cause and implemented a fix, with conversations returning to normal operation after 34 minutes, though some experienced higher-than-usual startup times during recovery.

Tavus experienced increased startup times for some conversations in their Conversational Video Interface for 18 minutes. The issue was identified and resolved by rerouting traffic to healthy providers. Normal service was restored after traffic routing adjustments were implemented.