Chalice-A4AB

What is this?

Chalice plugin: Support Agents for Amazon Bedrock

URL Links

PYPI :: Chalice-a4ab
https://pypi.org/project/chalice-a4ab/

Github :: Chalice-a4ab
https://github.com/ShotaOki/ChaliceA4AB

Usage

1. Install

   pip install -U chalice chalice-a4ab pydantic
   

2. Replace from chalice import Chalice to from chalice_a4ab import Chalice.

   Before:

   from chalice import Chalice
   
   app = Chalice("app-name")
   
   @app.router("path-name")
   ...

   After:

   from chalice_a4ab import Chalice
   
   app = Chalice("app-name")
   
   @app.router("path-name")
   ...

3. Update requirements.txt

   chalice-a4ab
   pydantic

4. Read parameters in function

   from chalice_a4ab import read_session_attributes, read_prompt_session_attributes, read_body
   
   @app.router("path-name")
   def method_name():
   
      # Read body
      body: PyDanticParserClass = read_body(app, PyDanticParserClass)
   
      # Read session attributes :: key -> session attribute key
      session_attributes: dict = read_session_attributes(app, "KEY")
   
      # Read prompt session attributes :: key -> prompt session attribute key
      prompt_session_attributes: dict = read_prompt_session_attributes(app, "KEY")
   
      # Return response (dict type)
      return {
         "response": "object"
      }

   or read directory

   @app.router("path-name")
   def method_name():
      # Read body
      body : dict = app.current_request.json_body
   
      # Read Session attributes or Prompt attributes from header
      for k in app.current_request.headers:
         if k.startsWith("sessionAttributes")
            # Session attributes
            print(app.current_request.headers[k])
         if k.startsWith("promptSessionAttributes")
            # Prompt session attributes
            print(app.current_request.headers[k])
   
      return {}

5. (Optional) Add Parser Lambda Function

   Add these decorator, function becomes "parser lambda"

   from chalice_a4ab import (
       Chalice,
       AgentsForAmazonBedrockConfig,
       ParserLambdaAbortException,
       ParserLambdaResponseModel,
   )
   
   # Set Config for Agents for Amazon bedrock
   AgentsForAmazonBedrockConfig().apply()
   
   ...
   
   # PRE_PROCESSING
   @app.parser_lambda_pre_processing()
   def pre_processing(event, default_result: ParserLambdaResponseModel) -> ParserLambdaResponseModel:
       # [MEMO] is_valid_input :: Allow/Deny API invocation
       # default_result.pre_processing_parsed_response.is_valid_input = True
       return default_result
   
   
   # ORCHESTRATION
   @app.parser_lambda_orchestration()
   def orchestration(event: dict, default_result: ParserLambdaResponseModel) -> ParserLambdaResponseModel:
       # [MEMO] Throw this Exception, Overwrite LLM response
       # raise ParserLambdaAbortException(message="Overwrited")
       return default_result
   
   # POST_PROCESSING
   @app.parser_lambda_post_processing()
   def post_processing(event, default_result: ParserLambdaResponseModel) -> ParserLambdaResponseModel:
      # [Memo] default: post_processing is invalid
      return default_result

6. Application works ::

   What you can do	Status
   Execute from Agents for Amazon Bedrock	〇
   Auto generate OpenAPI schema	×
   Management on cli	×

Advanced Usage

Create OpenAPI Schema automatically.

1. Install Chalice-a4ab and Chalice-spec

   pip install -U chalice chalice-spec==0.7.0 chalice-a4ab boto3 pydantic

2. Update requirements.txt

   chalice-spec==0.7.0
   chalice-a4ab
   pydantic

3. Add Setting

   from chalice_a4ab import Chalice, AgentsForAmazonBedrockConfig, ModelTypesAntropicClaudeV2
   from chalice_spec.docs import Docs, Operation
   
   # Set Config for Agents for Amazon bedrock
   AgentsForAmazonBedrockConfig(
      # Agent Situation
      instructions="Situation Settings for talking with Human and agent.(more than 40 words)",
      # Agent Description
      description="Description of application",
      # Use Model :: Claude V2
      foundation_model=ModelTypesAntropicClaudeV2,
   ).apply()
   
   app = Chalice(app_name="app-name")
   
   @app.router("path-name",
       methods=["POST"],
       docs=Docs(
           post=Operation(
               request=PyDanticRequestModelClass,
               response=PyDanticOutputModelClass,
           )
       ))
   def post_method():
      """
      Function summary
   
      Description of this method, brabrarba, this comment becomes LLM Prompt.
      """
   ...

4. Read values with PyDantic

   @app.router("path-name",
       methods=["POST"],
       docs=Docs(
           post=Operation(
               request=PyDanticRequestModelClass,
               response=PyDanticOutputModelClass,
           )
       ))
   def method_name():
      """
      Function summary
   
      Description of this method, brabrarba, this comment becomes LLM Prompt.
      """
   
      # Read body with utility
      body: PyDanticRequestModelClass = read_body(app, PyDanticRequestModelClass)
   
      # or direct
      #   Pydantic V1 -> parse_obj()
      #   Pydantic V2 -> model_validate()
      body = PyDanticRequestModelClass.model_validate(app.current_request.json_body)
   
      # return response as dict class
      #   Pydantic V1 -> json()
      #   Pydantic V2 -> model_dump_json()
      return PyDanticOutputModelClass(
         response=...
      ).model_dump_json()

   documentation for @app.router sample: https://github.com/TestBoxLab/chalice-spec

5. Management CLI

   Init command :: Create AWS Resource and OpenAPI Schema.

   chalice-a4ab init --profile ${PROFILE_NAME} --region ${REGION_NAME}

   Sync command :: Update Already AWS Resource and OpenAPISchema.

   chalice-a4ab sync --profile ${PROFILE_NAME} --region ${REGION_NAME}

   Show command :: Show OpenAPI Schema.

   chalice-a4ab show --profile ${PROFILE_NAME} --region ${REGION_NAME}

   Info command :: Show about Agents for Amazon Bedrock.

   chalice-a4ab info --profile ${PROFILE_NAME} --region ${REGION_NAME}

   Invoke command :: Invoke Agents for Amazon Bedrock.

   chalice-a4ab invoke "Natural Language Query" \
   --agent-id ${AGENT_ID} \
   --agent-alias-id ${AGENT_ALIAS_ID} \
   --end-session \
   --profile ${PROFILE_NAME} --region ${REGION_NAME}

   Delete AWS Resource

   chalice-a4ab delete --profile ${PROFILE_NAME} --region ${REGION_NAME}

6. Application works ::

   What you can do	Status
   Execute from Agents for Amazon Bedrock	〇
   Auto generate OpenAPI schema	〇
   Management on cli	〇

Develop

1. Setup

   poetry install

2. Run test

   poetry run pytest

Lisence

MIT

API

Construct

AgentsForAmazonBedrockConfig

Property	Descritpion
title	Application Name
instructions	Situation
foundation_model	Bedrock Model ID (Default: ModelTypesAntropicClaudeV2)
openapi_version	OpenAPI version (Support : 3.0 higher)
idle_session_ttl_in_seconds	Session timeout

Command Line TOOL

Command	Descritpion
chalice-a4ab init	Create AWS resource for Agents for amazon bedrock
chalice-a4ab sync	Sync OpenAPI schema to AWS
chalice-a4ab delete	Delete AWS resource for Agents for amazon bedrock

Options	Description
--bucket	Set S3 bucket name (for put OpenAPI schema)
--profile	Set AWS Profile Name
--region	Set AWS Region Name
--help	Show Help

API

AgentsForAmazonBedrockConfig

Method	Type	Description
apply	-	Current instace becomes global variable.
agents_for_bedrock_schema_json	-	Get OpenAPI Schema
save_schema_to_local	-	Save OpenAPI Schema to local folder
save_schema_to_s3	-	Upload OpenAPI Schema to S3 bucket
save_config_to_local	-	Save Config setting to local folder
get_global_config	static	Get global variable.