[bug] Remote inference with guardrails-ai==0.6.6 throws 401 error in deployed environment

[SurakshaPrabhu]

Description:
I’ve implemented and tested the code using guardrails-ai==0.6.6 locally, configured with my email-linked API token. Everything works as expected in the local environment.

However, when deployed in the remote environment, the remote inference requests fail with a 401 Unauthorized error. Below are the logs from the deployed environment:

DEBUG:urllib3.connectionpool:https://hub.api.guardrailsai.com:443 "POST /validator/restricttotopic/inference HTTP/1.1" 401 26
DEBUG:urllib3.connectionpool:https://hub.api.guardrailsai.com:443 "POST /validator/detect_jailbreak/inference HTTP/1.1" 401 26

What I've confirmed:

The same token and configuration work fine locally.

No changes were made to the token or validator setup between environments.

Token is correctly passed in the deployed code.

Expected Behavior:
Remote inference validators should work the same in the deployed environment as they do locally, using the same valid API token.

Request:
Please help identify if this is a token scope/authorization issue, an environment-specific configuration mismatch, or a backend issue with the Guardrails API.

[CalebCourier]

Hi @SurakshaPrabhu Can you please share details on how you are hosting you application? Specifically how it relates to how the guardrails package is configured during the build process and the the token being setup for use.

As long as the environment is properly configured, remote inference will work the same locally and in hosted environments. If your service is consistently receiving a 401 it is likely an issue of the token not being available in the hosted environment. If the 401 responses are inconsistent though, we can take a look to see if you are running into any other service limits that we can try to make more clear.

[Padmadas-aa]

@CalebCourier Thank you for reaching out.

We are currently hosting our application inside containers running on a Kubernetes platform (KPaas). During the build process, the guardrails package is installed and configured as part of our Docker image using a standard requirements.txt or pip install step.

As for authentication, the Guardrails token is injected at runtime via environment variables, which are securely managed through our Kubernetes secrets. This allows the application to authenticate with the Guardrails Hub during execution without hardcoding credentials in the image.