fix: send base64 images without data URL prefix to LM Studio

[roomote]

Description

This PR fixes an issue where LM Studio was receiving images with data URL prefixes (data:image/png;base64,) when it expects raw base64 encoded data only.

Problem

When using the Web Browser Tool with vision-capable models in LM Studio, users were encountering the error:
[Server Error] 'url' field must be a base64 encoded image

Solution

• Added an optional lmStudioFormat parameter to convertToOpenAiMessages function
• When enabled, this parameter strips the data URL prefix from images
• Updated the LM Studio handler to use this new parameter
• Maintains backward compatibility - default behavior unchanged for other providers

Testing

• Added comprehensive test coverage for both default and LM Studio formats
• All existing tests pass
• New test verifies the conditional behavior works correctly

Related Issue

Fixes #8827

Type of Change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

Checklist

• My code follows the style guidelines of this project
• I have performed a self-review of my code
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally with my changes

---

Important

Adds lmStudioFormat parameter to handle base64 images without data URL prefix for LM Studio in convertToOpenAiMessages.

• Behavior:
  • Adds lmStudioFormat parameter to convertToOpenAiMessages in openai-format.ts to strip data URL prefix from base64 images.
  • Updates LmStudioHandler in lm-studio.ts to use lmStudioFormat parameter.
  • Maintains backward compatibility; default behavior unchanged for other providers.
• Testing:
  • Adds tests in openai-format.spec.ts for both default and LM Studio formats.
  • Verifies conditional behavior for image data formatting.

^{This description was created by for 629830d. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Review Complete

No issues found. The implementation correctly handles LM Studio's requirement for raw base64 image data by:

• Adding an optional lmStudioFormat parameter to convertToOpenAiMessages
• Conditionally stripping the data URL prefix when enabled
• Maintaining backward compatibility for all other providers
• Including comprehensive test coverage for both formats

The fix is well-isolated and properly tested.

Follow Along on Roo Code Cloud

[roomote]

No issues found.

[MarkErik]

Now there is a different error in LM Studio: [Server Error] Invalid url

2025-10-25 19:10:07 [DEBUG]
Received request: POST to /v1/chat/completions with body {
"model": "qwen/qwen3-vl-30b",
"messages": [
{
"role": "system",
"content": "You are Roo, a highly skilled software engineer wi... ...user gives you instructions below to do otherwise."
},
{
"role": "user",
"content": [
{
"type": "text",
"text": "\nCan you check if my website at https://google.com is displaying correctly?\n"
},
{
"type": "text",
"text": "<environment_details>\n# VSCode Visible Files\n\n\n# V... ...or involves multiple steps.\n</environment_details>"
}
]
},
{
"role": "assistant",
"content": "<browser_action>\nlaunch\nhttps://google.com\n</browser_action>"
},
{
"role": "user",
"content": [
{
"type": "text",
"text": "[browser_action for 'launch'] Result:"
},
{
"type": "text",
"text": "The browser action has been executed. The console ... ...rowser before you can use the write_to_file tool.)"
},
{
"type": "image_url",
"image_url": {
"url": "UklGRt5aAABXRUJQVlA4WAoAAAAgAAAAgwMAVwIASUNDUMgBAA... ...p5IKVQI3kvAwpRJ8MxVMmXCBzK/ym+PVWjcCCVIAAAAAAAAAA="
}
},
{
"type": "text",
"text": "<environment_details>\n# VSCode Visible Files\n\n\n# V... ...or involves multiple steps.\n</environment_details>"
}
]
}
],
"temperature": 0,
"stream": true
}
2025-10-25 19:10:07 [ERROR]
[Server Error] Invalid url

[MarkErik]

I have tracked down the issue - LMStudio wants images in PNG format, not webP.

I'll submit a PR with a fix for this.