☁️ 6.2 Bedrock Runtime

A MATLAB client for the Amazon Bedrock Runtime service. It provides API operations for running inference against Bedrock foundation models.

bedrock = aws.bedrock.runtime.Client('region', 'us-east-1');

Note

Bedrock Runtime is not available in some default regions (e.g., us-west-1). Use a supported region such as us-east-1.

🔧 6.2.1 List of Available Methods

🧩 6.2.2 Examples

🔸 Invoke a Model (Text Generation)

textPrompt = "What is your name?";

response = bedrock.invokeModel( ...
    modelId = "amazon.titan-text-express-v1", ...
    body    = textPrompt);

disp(response.outputText);

Note

Some models provided in the example might have reached end of life. Use the updated model_id in case of errors like “This model version has reached the end of its life”.

🔸 Invoke a Model (Image Generation)

spec = struct(text="Neon skyline at dusk", seed=42, style="photographic");
resp = bedrock.invokeModel(modelId="amazon.titan-image-generator-v1", body=spec);
disp(resp.responseBody);

🔸 Stateful Conversation (Multi-Turn; Text)

firstPrompt = "Tell me about yourself in 50 words?";
message1 = aws.bedrock.runtime.model.Message(text = firstPrompt, role = "user");

fullConversationMessages = aws.bedrock.runtime.model.Message.empty;
fullConversationMessages(end+1) = message1;

response1 = bedrock.converse( ...
    modelId  = "amazon.titan-text-lite-v1", ...
    messages = fullConversationMessages);

fullConversationMessages(end+1) = response1.message; % keep context

secondPrompt = "Can you elaborate to 600 words?";
message2 = aws.bedrock.runtime.model.Message(text = secondPrompt);
fullConversationMessages(end+1) = message2;

response2 = bedrock.converse( ...
    modelId   = "amazon.titan-text-lite-v1", ...
    messages  = fullConversationMessages, ...
    maxTokens = 600);

disp(response2.message.text);

🔸 Converse with Image (Vision Models Only)

imgMsg = aws.bedrock.runtime.model.Message(imageData = imageArray); % numeric image

resp = bedrock.converse( ...
    modelId  = "<vision-capable-model-id>", ...
    messages = imgMsg);

disp(resp.message.text);

📘 6.2.3 Method Reference (Summary)

🔸 `invokeModel`

response = bedrock.invokeModel(modelId = "<model-id>", body = <string>);

Returns: aws.bedrock.runtime.model.InvokeModelResponse
- outputText, completionReason, tokenCount (optional)

🔸 `converse`

response = bedrock.converse( ...
    modelId   = "<model-id>", ...
    messages  = <Message | Message[]>, ...
    maxTokens = <int>);  % optional

Returns: aws.bedrock.runtime.model.ConverseResponse
- message (assistant message), stopReason, usage

💡 Notes & Best Practices

Prefer name-value argument style for clarity and forward compatibility.
Maintain the conversation array by appending both user and assistant turns to preserve context.
Use a supported region like us-east-1 for Bedrock Runtime availability.
Roles supported here are user and assistant.
Pass a scalar string for quick text prompts. Provide a struct/dictionary (or JSON string) when the model requires additional parameters (for example, image generation seed or style).