Upgraded to 0.8.3 rc1 and received errors about agents and tools enabled

[matp-zelus]

What happened?

Used Helm to upgrade from 0.8.2 to 0.8.3-rc1 in a GKE Autopilot Cluster, we are using Vertex AI via auth.json. After logging in to LibreChat it generated an error when a prompt was submitting stating it was not possible to have Agents enabled and also tools enabled - this wasn't an error previously (but may be due to a misconfiguration so I've shared the extract from modelSpecs below).

It seems to have been caused by either webSearch or web_search which used to work without issue under modelSpecs - can do some more investigation to narrow it down if necessary:

  list:
    - name: "Gemini 2.5 Flash"
      label: "Fast Thinking"
      executeCode: false
      webSearch: true
      description: "Fast Thinking with Google Search"
      default: true
      preset:
        endpoint: "google"
        model: "gemini-2.5-flash"
        web_search: true

...

interface:
  agents: true

Version Information

have pinned to 1.9.7 (0.8.2) for the moment:

echo "Deploying via Helm..."
helm upgrade --install "librechat"
oci://ghcr.io/danny-avila/librechat-chart/librechat
--version 1.9.7
--values "$VALUES_FILE"
--timeout 20m
--wait

Steps to Reproduce

1. Used Helm to upgrade from 0.8.2 to 0.8.3-rc1 in a GKE Autopilot Cluster
2. Login to LibreChat
3. Submit a prompt
4. Receive error message immediately

What browsers are you seeing the problem on?

Chrome

Relevant log output

Can provide more info if necessary but will need to do it later when users aren't on the platform.

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Switch this to false, it’s for the LibreChat native web search. Google does not allow using their native tools (like grounding/search) along with other tools.

webSearch: true

A bug was fixed with model specs where these “badge” settings weren’t always being applied. You had a misconfiguration which was not being reflected at runtime.

[matp-zelus]

Thanks very much Danny, that's working perfectly now, and thanks for your stellar work on Librechat, I've had a fantastic time learning how to deploy and configure LibreChat in GKE using an autopilot cluster - it's been challenging and I've learnt a lot. I was really interested to read your Roadmap last week - great to see where you're going with it. I've started looking into the McpServers today, great to see the extended config released last week.

I'm really looking forward to the general availability of the Code Interpreter as Gemini / Vertex AI don't allow you to process MS Office files (as they are effectively zip files and they don't provide a sandbox like OpenAI do) and I'm hoping I can do so using Python within LibreChat. Plus the admin dashboard sounds like it will be a massive timesaver, although I'm using OpenID so it might not be aimed at my use case.

I do have another question about image generation but can see a separate request for that so I'll add to that. Thanks again for all your hard work.