ask-bedrock 0.0.6

Converse with your favorite Amazon Bedrock LLM from the command line.

Ask Amazon Bedrock

Converse with your favorite Amazon Bedrock large language model from the command line.

This tool is a wrapper around the low-level Amazon Bedrock APIs and Langchain. Its main added value is that it locally persists AWS account and model configuration to enable quick and easy interaction.

Installation

⚠️ Requires Python >= 3.9

⚠️ Requires a working AWS CLI setup configured with a profile that allows Amazon Bedrock access. See CLI documentation for details.

pip install ask-bedrock

You can also build/run this project locally, see Building and Running Locally.

Usage

Activating models

Before you can use this command line tool, you need to request model access through the AWS Console in a region where Bedrock is available: Switch to the region where you want to run Bedrock, go to ”Model access“, click “Edit”, activate the models you wish to use, and then click “Save changes”.

Invocation

To start a conversation, simply enter the following command:

ask-bedrock converse

Upon the first run, you will be led through a configuration flow. To learn more about configuration options, see the Configuration section below.

If you’re fully configured, the tool will show you a >>> prompt and you can start interacting with the configured model.

Multi-line prompts can be wrapped into <<< >>> blocks.

To end your interaction, hit Ctrl + D. Note that the conversation will be lost.

Configuration

Ask Amazon Bedrock stores your user configuration in $HOME/.config/ask-bedrock/config.yaml. This file may contain several sets of configuration (contexts). For instance, you can use contexts to switch between different models. Use the --context parameter to select the context you'd like to use. The default context is default.

If no configuration is found for a selected context, a new one is created. If you want to change an existing config, use

ask-bedrock configure --context mycontext

You can also create or edit the configuration file yourself in $HOME/.config/ask-bedrock/config.yaml:

contexts:
  default:
    region: ""                  # an AWS region where you have activated Bedrock
    aws_profile: ""             # a profile from your ~/.aws/config file
    model_id: ""                # a Bedrock model, e.g. "ai21.j2-ultra-v1"
    model_params: "{}"          # a JSON object with parameters for the selected model

Model parameters

This JSON is passed to Langchain during client setup (as model_kwargs). The schema depends on the model that is used. Have a look at the examples.

If you want to configure multiple lines, model parameters can be wrapped in <<< >>>.

Building and Running Locally

pip install build
python -m build
python ask_bedrock/main.py converse

Feedback

As this tool is still early stage, we are very interested in hearing about your experience. Please take one minute to take a little survey: https://pulse.aws/survey/GTRWNHT1

Troubleshooting

Q: I’m getting the following error during invocation: “ValueError: Error raised by bedrock service: An error occurred (AccessDeniedException) when calling the InvokeModel operation: Your account is not authorized to invoke this API operation.”

A: You may have selected a model that is currently not yet activated for public usage. It may have been listed it in the selection of available models, but unfortunately some models (such as Amazon Titan) aren’t yet available via API.

---

Q: The model responses are cut off mid-sentence.

A: Configure the model to allow for longer response. Use model parameters (see above) for this. Claude for example would take the following model parameters: {"max_tokens_to_sample": 3000}

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.