Expand core api to allow for reading images

[bdougie]

I am working on an agent using the api - https://github.com/bdougie/vision

A lot of the work in my agent is preparing a video for using the vision model to return a description. Today the api doesn't quite send the image to ollama for processing.

https://github.com/bdougie/vision/blob/51fb3a17f0b7e9273798c05f86ca435aa575d109/main.go#L41-L59

func analyzeImage(ctx context.Context, a *agent.DefaultAgent, imagePath string) (string, error) {
	imageData, err := os.ReadFile(imagePath)
	if (err != nil) {
		return "", err
	}


	// Create vision prompt with image data
	prompt := fmt.Sprintf(`[
		{"type": "text", "text": "Describe this image in detail."},
		{"type": "image", "source": {"data": "%s", "media_type": "image/jpeg"}}
	]`, imageData)


	response, err := a.Run(ctx, prompt, agent.DefaultStopCondition)
	if err != nil {
		return "", err
	}

        // this line does not return what I need today. 
	return response[0].Message.Content, nil
}

I may be wrong on this as I am still figuring it out, but the ollama way locally, I can simply to do this:

ollama run llama3.2-vision:11b
>>> describe the image at this location /./frame_0003.jpg

What would I like to see?

Perhaps a tool that is ready to manage an image for the model.

[jpmcb]

Currently core doesn't really support images although they exist on the "image" type:

// Message represents a single message in a conversation with multimodal support
type Message struct {
        // etc. etc.

	// A list of base64-encoded images (for multimodal models such as llava
	// or llama3.2-vision)
	Images []string
}

It should be as simple as getting the base64 encodings for the images and passing it to the agent. Something like:

func (a *Agent) WithImages(base64Img string) 

A few things to think through:

• Do we expect that an end user would already have the images base64 encoded? Or should that be something that agent-api does for you?