Tavus had 48 outages in the last 2 years totaling 27h 16m of downtime — averaging 2 incidents per month.
There were 48 Tavus outages since August 22, 2025 totaling 27h 16m of downtime. Each is summarised below — incident details, duration, and resolution information.
We are investigating an elevated error rate in image-to-face trainings.
A fix has been rolled out and we are seeing errors resolve. We are continuing to monitor the fix
This incident has been resolved.
When one of our GPU providers performed maintenance, we asked to migrate traffic to a different cluster beforehand to avoid disruptions. During the migration back, the provider encountered an issue that led to some conversations experiencing an elevated startup time.
We are investigating an issue affecting a small number of conversations configured to use Cartesia for TTS. PAL may intermittently fail to respond during conversations. We are actively investigating the cause.
We are working with Cartesia to restore normal service as quickly as possible. Upstream status: https://status.cartesia.ai/
Cartesia has restored service, and conversations using Cartesia TTS are responding normally again. We are actively monitoring the service to ensure continued stability.
Cartesia has fully restored service, and conversations using Cartesia TTS are responding normally.
We are experiencing some custom replica training failures due to an upstream infra provider issue. We are working on restarting all new customer replica trainings as soon as possible. Nothing is needed from your side.
The upstream provider has recovered and we're relaunching custom replica trainings.
This incident has been resolved.
An outage affecting our upstream TTS provider, Cartesia, is causing elevated error rates for some conversations where Cartesia is the default TTS and causing some to fail to start.
One of our GPU providers is experiencing networking issues, which is impacting conversations configured to use Cartesia for TTS and may cause them to fail to start. We are actively routing traffic away from the affected region
All traffic has been routed away from the affected provider, and conversations are starting normally. We are actively monitoring the service to confirm continued stability.
All traffic has been routed away from the affected provider, and conversations are starting normally.
An outage affecting our upstream TTS provider, Cartesia, is causing some conversations using Cartesia as the default TTS to fail to start. Elevenlabs users are not affected
https://status.cartesia.ai/ Cartesia is failing over traffic. Elevnlabs users are not affected
We are seeing traffic return to normal
This incident has been resolved.
We are investigating increased conversation latency affecting conversations using the Gemma 4 model. Conversations using other models are operating normally. We are actively working to identify the cause and will provide an update as soon as more information is available.
Our upstream model provider is rate- limiting a subset of requests for the Gemma 4 model, resulting in increased latency. We are routing traffic away from the affected capacity and working with the provider to restore normal performance.
Conversation latency for the Gemma 4 model has returned to normal, and all traffic has been routed.
We are currently investigating the issue
We have identified the issue. An outage affecting our upstream TTS provider, Cartesia, is causing some conversations using Cartesia as the default TTS to fail to start. Only conversations configured to use Cartesia for TTS are affected. Cartesia is working on resolving this issue, and we're continuing to monitor. https://status.cartesia.ai/incidents/01KYQZ97RPSZBJRPEQGVGJANZP
We are working with Cartesia to restore service
Cartesia has restored service, and conversations configured to use Cartesia for TTS are starting successfully again.
We are investigating an issue causing some conversations to fail to start.
tavus-oss conversational LLM rate-limited after an upstream provider change. Latency and dropped turns possible.
Uncovered and rolled back related change - OSS LLM (tavus-oss) traffic is healthy again.
We are currently investigating this issue.
A fix has been implemented and we are monitoring the results.
This incident has been resolved.
We have identified an issue where newly created replicas are failing to join CVI conversations. Existing replicas are unaffected. This issue is caused by a problem with a downstream provider, who has been notified and is actively working on a resolution. We will provide further updates as the situation develops.
A fix has been deployed and we have identified the root cause. Only a very limited subset of conversations using newly trained replicas were affected by join failures. We are continuing to monitor the system.
A fix has been deployed and this incident is now resolved. A very limited subset of conversations using newly trained replicas were affected by join failures. We identified the root cause and do not expect any further impact.
Async video generation and replica training are encountering slower than expected processing times. While requests are completing successfully, they are taking longer than expected. We have identified a downstream provider encountering hardware slow downs. We are working with the provider to expedite a resolution. This incident does not impact CVI.
This issue has been resolved, and all existing replica trainings and video generations are now processing as expected. Our team is continuing to monitor all runs.
We are investigating an infrastructure issue that is causing some conversations not to start
A fix has been implemented and we are monitoring the results.
This incident has been resolved.
We’re currently seeing elevated error rates across multiple parts of our stack, including perception, knowledge base, Gemini conversation LLM, objectives, and guardrails.
We’ve identified the root cause of the issue and are working on deploying a fix. Update: All services, except perception, are currently operating as expected.
All deployments have been restored, and the issue with our cloud provider has been resolved. Services are now up and running, and we’re continuing to monitor to ensure stability.
This incident has been resolved.
We are investigating increased error rates with Raven-0 and Raven-1
The issue was identified and a fix has been deployed
Raven-0 and Raven-1 are now operating normally, and we are actively monitoring the system to ensure continued stability.
The incident has been resolved, and Raven is now working as expected.
There are intermittent call connection issues tied to upstream Daily and Cloudflare outages. Daily: https://status.daily.co/ Cloudflare: https://www.cloudflarestatus.com/
Cloudflare has implemented a fix and downstream services are returning to normal service
We are currently investigating this issue.
This has been resolved.
We are seeing elevated API errors from our webRTC provider Daily https://status.daily.co/incidents/0vkhkxd7m20b
A fix has been released by the Daily team, we are seeing traffic return to normal and will be continuing to monitor
This incident has been resolved
Conversations requests are succeeding, but encountering issues at the GPU level which are causing replicas to not properly join
A sub GPU provider has encountered a 10 minute outage involving their AWS instance. Traffic was automatically rerouted, although some conversations which were already sent to this provider did not properly join. All traffic is now performing within SLA, and we're continuing to monitor.
This incident has been resolved.
An underlying GPU provider is encountering queuing. We are actively switching traffic Note this is not impacting all traffic.
Traffic has been shifted away from this provider and we're continuing to monitor
This incident has been resolved.
We are currently investigating
We’ve identified the root cause of the issue and are actively monitoring the conversations.
We are continuing to monitor for any further issues.
Conversations are now starting normally. Some conversations may still experience higher-than-usual startup times. We’re continuing to monitor the situation closely.
We experienced a brief disruption due to a configuration change that impacted conversation routing, causing some conversations to stall or take longer to start. The issue has been resolved and all systems are now operating normally.
The issue has been identified and a fix is being implemented.
We have rerouted traffic to healthy providers. Traffic has returned to normal.
This incident has been resolved.
We detected several conversations with start times exceeding 10 seconds. We have diverted traffic and are continuing to monitor the situation.
We have identified an issue with one of our GPU providers and shifted traffic away from them