Image Processing Fails with Gemma3 Model via Ollama (Khoj 1.38.0)

[consulitsk]

Server

• Cloud (https://app.khoj.dev)
• Self-Hosted Docker
• Self-Hosted Python package
• Self-Hosted source code

Clients

• Web browser
• Desktop/mobile app
• Obsidian
• Emacs
• WhatsApp

OS

• Windows
• macOS
• Linux
• Android
• iOS

Khoj version

1.38.0

Describe the bug

We encountered an issue when using the Khoj 1.38.0 version with the Gemma3 model via Ollama, with vision enabled.

Current Behavior

When an image is submitted, the following errors appear in the logs:

khoj.routers.storage: AWS is not enabled. Skipping image upload
khoj.processor.conversation.openai.utils Error code: 400 -
{'error': {'message': 'invalid image input', 'type': 'invalid_request_error', 'param': None, 'code': None}}

Based on the following issue, this should have been fixed already:
#1112

Is there any additional configuration required to properly handle image inputs with vision enabled? Or could this be a regression?

Expected Behavior

Image inputs should be processed correctly when using the Gemma3 model via Ollama with vision enabled. The image should be accepted and analyzed without errors, and no AWS-related upload warnings should appear if cloud storage is not configured or required.

Reproduction Steps

Run Khoj version 1.38.0.
Use the Gemma3 model via Ollama with vision enabled.
Upload an image in a conversation.

Possible Workaround

No response

Additional Information

No response

Link to Discord or Github discussion

No response