Skip to main content
Local models can take longer to generate a response, especially through a tunnel or gateway. Set AI_TIMEOUT_MULTIPLIER to give every web app AI operation more time. It accepts numbers from 1 to 10, including decimals such as 1.5. The default is 1; leaving it unset or blank preserves today’s limits. Invalid values prevent the app from starting. The call budget includes retries. QSF attempts share their call’s budget and are also limited by the remaining import deadline; five seconds remain reserved for assembling the draft. The setting applies to every AI provider configured for the web app. Hub and taxonomy have their own configuration, and OAuth token requests retain their separate ten-second timeout.
1

Set the multiplier

For Docker Compose, add this to docker/.env:
For Helm, set the application environment in your values:
2

Apply the environment change

From the Docker Compose directory, recreate the app container so it reads the updated environment:
For Kubernetes, apply your Helm values with your usual upgrade command and let the app pods restart.
3

Verify a slow operation

Generate a survey or translate one against your model and confirm it finishes. Pressing Stop during survey generation still cancels the model request immediately. To restore the default limits, set the multiplier to 1 and apply the environment change again.
Your ingress, reverse tunnel, gateway, and model server can impose shorter limits. Increase those limits as needed: non-streaming operations may send no response until generation finishes. Survey generation and QSF imports send a keepalive after ten seconds of silence, but that does not override a proxy’s maximum request duration or a provider’s own timeout.