Summary

Implemented comprehensive vision support for multimodal messages to enable requests to vision-capable models like GPT-4 Vision, along with backward compatibility improvements and example refinements.

Changes

Vision Support

• Updated OpenAPI spec to latest version with vision types
• Generated new types: ContentPart, ImageContentPart, TextContentPart, ImageURL
• Added helper functions for easy message creation:
  • NewTextMessage(role, text) - Create simple text messages
  • NewImageMessage(role, parts) - Create multimodal messages with images
  • NewTextContentPart(text) - Create text content parts
  • NewImageContentPart(imageURL, detail) - Create image content parts
• Comprehensive test coverage with vision_test.go
• Added vision example in examples/vision/main.go
• Updated README with vision documentation

Backward Compatibility

• Added NewMessageContent[T string | []ContentPart](value T) generic helper
  • Uses Go generics to accept both string and []ContentPart
  • Follows OpenAI SDK pattern for intuitive API
  • Example: Content: sdk.NewMessageContent("Hello world")
• Updated all tests and examples to use the new helper
• Renamed NewVisionMessage to NewImageMessage for better clarity

Example Improvements

• Refactored models example to use provider loop instead of repetitive code
• Fixed context timeout issues by creating fresh contexts for each request
• Reduced models example from ~80 lines to ~40 lines
• Proper context cleanup with cancel() calls

Other Changes

• Removed A2A-related middleware (replaced with MCP)
• Updated middleware options to use SkipMCP and DirectProvider
• Maintained backward compatibility with existing code patterns

Breaking Changes

⚠️ Message.Content is now Message_Content union type instead of string

Migration:

// Old (broken):
Message{Role: User, Content: "Hello"}

// New (works):
Message{Role: User, Content: sdk.NewMessageContent("Hello")}

// Or use the helper:
msg, err := sdk.NewTextMessage(sdk.User, "Hello")

Supported Features

• ✅ Image URLs (https://)
• ✅ Base64-encoded images (data:image/...)
• ✅ Multiple images per message
• ✅ Image detail levels (auto, low, high)
• ✅ String content (via NewMessageContent)
• ✅ Multimodal content arrays

Testing

All tests pass successfully:

task test  # All tests pass

All examples compile and run:

• examples/generation/ - Basic text generation
• examples/stream/ - Streaming responses
• examples/tools/ - Function calling
• examples/vision/ - Vision/multimodal messages
• examples/models/ - List provider models
• examples/middleware-bypass/ - Middleware control

Closes #26

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com