module pixeltable.functions.bedrock

Pixeltable UDFs for AWS Bedrock AI models. In order to use them, you must first configure your AWS credentials, as described in the Working with Bedrock tutorial.

func invoke_tools()

Signature

invoke_tools(
    tools: pixeltable.func.tools.Tools,
    response: pixeltable.exprs.expr.Expr
) -> pixeltable.exprs.inline_expr.InlineDict

Converts a Bedrock response dict to Pixeltable tool invocation format and calls tools._invoke().

udf converse()

Signature

@pxt.udf
converse(
    messages: pxt.Json[(Json, ...)],
    *,
    model_id: pxt.String,
    system: pxt.Json[(Json, ...)] | None = None,
    inference_config: pxt.Json | None = None,
    additional_model_request_fields: pxt.Json | None = None,
    tool_config: pxt.Json[(Json, ...)] | None = None
) -> pxt.Json

Generate a conversation response. Equivalent to the AWS Bedrock converse API endpoint. For additional details, see: https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime/client/converse.html PIL images and media file paths in messages[*].content[*].(image|video|audio).source.bytes are converted to raw bytes automatically. Parameters:

messages (pxt.Json[(Json): Input messages.
model_id (Any): The model that will complete your prompt.
system (Any): An optional system prompt.
inference_config (Any): Base inference parameters to use.
additional_model_request_fields (Any): Additional inference parameters to use.
tool_config (Any): An optional list of Pixeltable tools to use.

Returns:

pxt.Json: A dictionary containing the response and other metadata.

Examples: Add a computed column that applies the model us.anthropic.claude-haiku-5-5 to an existing Pixeltable column tbl.prompt of the table tbl:

msgs = [{'role': 'user', 'content': [{'text': tbl.prompt}]}]
tbl.add_computed_column(
    response=converse(msgs, model_id='us.anthropic.claude-haiku-5-5')
)

Pass an image via the Converse API:

msgs = [
    {
        'role': 'user',
        'content': [
            {'image': {'format': 'jpeg', 'source': {'bytes': tbl.image}}},
            {'text': "What's in this image?"},
        ],
    }
]
tbl.add_computed_column(
    response=converse(msgs, model_id='amazon.nova-lite-v1:0')
)

udf embed()

Signatures

# Signature 1:
@pxt.udf
embed(
    text: pxt.String,
    *,
    model_id: pxt.String,
    dimensions: pxt.Int | None
) -> pxt.Array[(None,), float32]

# Signature 2:
@pxt.udf
embed(
    image: pxt.Image,
    *,
    model_id: pxt.String,
    dimensions: pxt.Int | None
) -> pxt.Array[(None,), float32]

Generate text or image embeddings using Amazon Titan, Amazon Nova, or Cohere embedding models. Calls the AWS Bedrock invoke_model API for embedding models. For additional details, see: https://docs.aws.amazon.com/bedrock/latest/userguide/titan-embedding-models.html https://docs.aws.amazon.com/nova/latest/userguide/modality-embedding.html https://docs.aws.amazon.com/bedrock/latest/userguide/model-parameters-embed.html Parameters:

text (String): Input text to embed.
model_id (String): The embedding model identifier. Supported models:
- amazon.titan-embed-text-v1
- amazon.titan-embed-text-v2:0 (supports dimensions: 256, 512, 1024)
- amazon.nova-2-multimodal-embeddings-v1:0 (supports dimensions: 256, 512, 1024, 3072)
- cohere.embed-english-v3
- cohere.embed-multilingual-v3
- cohere.embed-v4:0 (supports dimensions: 256, 512, 1024, 1536)
dimensions (Int | None, default: Literal(None)): Output embedding dimensions (model-dependent, optional).

Returns:

pxt.Array[(None,), float32]: Embedding vector.

Examples: Create an embedding index on a column description with Nova embeddings and custom dimensions:

tbl.add_embedding_index(
    tbl.description,
    string_embed=embed.using(
        model_id='amazon.nova-2-multimodal-embeddings-v1:0',
        dimensions=1024,
    ),
)

udf invoke_model()

Signature

@pxt.udf
invoke_model(body: pxt.Json, *, model_id: pxt.String) -> pxt.Json

Invoke a Bedrock model. Equivalent to the AWS Bedrock invoke_model API endpoint, with automatic routing to StartAsyncInvoke for models that require it (e.g. video generation, audio/video embeddings). For additional details, see: https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime/client/invoke_model.html PIL images and media file paths anywhere in the request body are converted automatically to the encoding expected by the target model. For image-generation models, base64-encoded images in the response are automatically decoded into PIL.Image objects. For models that require async invocation, bedrock.temp_location must be configured (set environment variable BEDROCK_TEMP_LOCATION or add temp_location to the [bedrock] section of your Pixeltable configuration file). Parameters:

body (pxt.Json): The prompt and inference parameters as a dictionary.
model_id (pxt.String): The model identifier to invoke.

Returns:

pxt.Json: A dictionary containing the model response. For image-generation models, image fields are decoded to PIL.Image objects. For video-generation models, returns a pxt.Video path.

Examples: Invoke Amazon Titan text embeddings:

body = {'inputText': tbl.text, 'dimensions': 512, 'normalize': True}
tbl.add_computed_column(
    response=invoke_model(body, model_id='amazon.titan-embed-text-v2:0')
)

Invoke TwelveLabs Marengo with an image column (note that the image can be included directly in the invoke body, and will be automatically base64-encoded by Pixeltable):

body = {
    'inputType': 'image',
    'image': {'mediaSource': {'base64String': tbl.image}},
}
tbl.add_computed_column(
    response=invoke_model(
        body, model_id='twelvelabs.marengo-embed-3-0-v1:0'
    )
)

Invoke TwelveLabs Marengo with audio (auto-routes to async via StartAsyncInvoke; requires bedrock.temp_location to be configured):

body = {
    'inputType': 'audio',
    'audio': {'mediaSource': {'base64String': tbl.audio}},
}
tbl.add_computed_column(
    response=invoke_model(
        body, model_id='twelvelabs.marengo-embed-3-0-v1:0'
    )
)

Invoke Anthropic Claude with an image:

body = {
    'anthropic_version': 'bedrock-2023-05-31',
    'max_tokens': 1024,
    'messages': [
        {
            'role': 'user',
            'content': [
                {
                    'type': 'image',
                    'source': {
                        'type': 'base64',
                        'media_type': 'image/jpeg',
                        'data': tbl.image,
                    },
                },
                {'type': 'text', 'text': "What's in this image?"},
            ],
        }
    ],
}
tbl.add_computed_column(
    response=invoke_model(body, model_id='us.anthropic.claude-haiku-5-5')
)

Invoke Amazon Nova Lite with a video column:

body = {
    'messages': [
        {
            'role': 'user',
            'content': [
                {
                    'video': {
                        'format': 'mp4',
                        'source': {'bytes': tbl.video},
                    }
                },
                {'text': 'What happens in this video?'},
            ],
        }
    ]
}
tbl.add_computed_column(
    response=invoke_model(body, model_id='amazon.nova-lite-v1:0')
)

Invoke Stability AI for image generation:

body = {
    'prompt': tbl.prompt,
    'mode': 'text-to-image',
    'aspect_ratio': '1:1',
    'output_format': 'jpeg',
}
tbl.add_computed_column(
    response=invoke_model(body, model_id='stability.sd3-5-large-v1:0')
)
tbl.add_computed_column(image=tbl.response['images'][0])

Invoke Luma Ray 2 for video generation (auto-routes to async; requires bedrock.temp_location). Luma Ray 2 (luma.ray-v2:0) is available in us-west-2:

body = {
    'prompt': tbl.prompt,
    'aspect_ratio': '16:9',
    'duration': '5s',
    'resolution': '540p',
}
tbl.add_computed_column(
    video=invoke_model(body, model_id='luma.ray-v2:0')
)

​module pixeltable.functions.bedrock

​func invoke_tools()

​udf converse()

​udf embed()

​udf invoke_model()

module pixeltable.functions.bedrock

func invoke_tools()

udf converse()

udf embed()

udf invoke_model()