Add Amazon Bedrock Agent Support #676

yosefkri · 2025-08-12T07:27:54Z

Overview

This PR adds support for Amazon Bedrock agents to the AWS GenAI LLM Chatbot. Users can now configure and use Bedrock agents through the chatbot interface, providing an additional AI capability alongside existing LLM providers.

Key Changes

Configuration Updates
- Extended the bedrock section in config.json to include agent configuration
- Added CLI prompts for Bedrock agent setup (agent ID, version, alias ID)
- Added support for non-interactive configuration via environment variables
Backend Implementation
- Created a new Bedrock agent client module in the Python SDK layer
- Implemented a Bedrock agent adapter that extends the BaseAdapter
- Added support for processing EventStream responses from Bedrock agents
- Implemented fallback mechanisms for response handling
Environment and Permissions
- Added environment variables for Bedrock agent configuration
- Updated IAM permissions to allow Lambda functions to invoke Bedrock agents
Model Registration
- Updated the direct model provider to include Bedrock agent models
- Registered the Bedrock agent adapter with the pattern "bedrock.bedrock_agent"
Error Handling and Logging
- Added specific error handling for Bedrock agent-related errors
- Enhanced logging for Bedrock agent interactions

Testing

Tested the configuration flow to ensure agent settings are properly saved
Verified integration with Bedrock agent using sample queries
Validated response handling from Bedrock agents
Tested error handling for various failure scenarios
Verified cross-region functionality works correctly

Challenges Addressed

Response Format: Implemented handling for the nested EventStream structure in Bedrock agent responses
Region Configuration: Added explicit region configuration for Bedrock agent clients
Event Processing: Added support for dictionary events in the EventStream

Documentation

Added implementation plan document with detailed architecture and design decisions
Updated relevant documentation to include Bedrock agent configuration and usage

How to Test

Configure a Bedrock agent in your AWS account

Update your config.json with the agent details:

"bedrock": {
  "enabled": true,
  "region": "your-region",
  "agent": {
    "enabled": true,
    "agentId": "your-agent-id",
    "agentVersion": "DRAFT",
    "agentAliasId": "your-agent-alias-id"
  }
}

Deploy the chatbot
Create an application using the "bedrock_agent" model
Test interactions with your Bedrock agent

Future Enhancements

Support for streaming responses from Bedrock agents
Enhanced error handling and recovery mechanisms
UI improvements for agent-specific features
Integration with agent action groups
Support for agent knowledge bases

* Adding support using bedrock agent * update langchain issue * updated plan readme --------- Co-authored-by: Yossi Kricheli <yossik@amazon.com>

maryamkhidir · 2025-08-12T08:05:26Z

lib/shared/layers/python-sdk/python/genai_core/bedrock_agent/client.py

+                        logger.info(f"Completion value type: {type(completion_value)}")
+                        # Try to extract the completion from the EventStream
+                        try:
+                            # Process the EventStream


Can you refactor this to reduce nesting and improve readability? I'd extract the response stream processing and extraction logics out into chunks.

Can you also add some unit tests to cover this?

maryamkhidir · 2025-08-12T08:08:57Z

lib/shared/layers/python-sdk/python/genai_core/model_providers/direct/provider.py

+        return [
+            {
+                "provider": Provider.BEDROCK.value,
+                "name": "bedrock_agent",


why are we not using the agent alias here, or including it in the name?

maryamkhidir

Thanks for this @yosefkri.
QQ: Will there be a follow up CR for the UI Improvements e.g add Agents trace to the chatbot console?

One observation: The current approach only allows for a single pre-configured agent in the chatbot. For better flexibility, we may want to consider supporting multiple agents. Perhaps we could explore retrieving all available agents from the customer's Bedrock environment?

For reference, this CR provides some insights:

aws-genai-llm-chatbot/lib/shared/layers/python-sdk/python/genai_core/models.py

Line 41 in a1c44fd

def list_bedrock_agents():

* Adding support using bedrock agent * update langchain issue * updated plan readme * Making code more readable --------- Co-authored-by: Yossi Kricheli <yossik@amazon.com>

Rob-Powell

I think this file ‘bedrock-agent-implementation-plan.md’ should be in the ./docs directory of the project somewhere not in the root.
Your PR also mentions how you've updated relevant docs but I don't see any of those yet. Is this PR still a WIP?

yosefkri · 2025-08-13T06:49:49Z

@Rob-Powell the PR is WIP since we decided to move to discover bedrock agent dynamically instead using specific agent ID.
The plan needs to be removed, I will update the documentation also.

* Adding support using bedrock agent * update langchain issue * updated plan readme * Making code more readable * Added list agent feature * Added support to specific agent or list agents * Added support for bedrock agent by passing the s3 uri for file upload --------- Co-authored-by: Yossi Kricheli <yossik@amazon.com>

yosefkri · 2025-08-21T11:52:58Z

Pull Request Summary: Enhanced Bedrock Agent Support

This pull request adds comprehensive support for Amazon Bedrock Agents in the AWS GenAI LLM Chatbot, with two main components:

1. S3 URI-Based File Handling for Bedrock Agents

We've implemented a more efficient approach to file handling with Bedrock Agents by providing S3 URIs instead of embedding file content directly in prompts:

Removed base64 encoding of file content, eliminating token consumption and size limitations
Enhanced prompt structure with clear sections for file information, user queries, and S3 access details
Preserved original file keys throughout the processing pipeline to ensure correct S3 paths
Added conditional file information that only includes file handling instructions when files are present
Improved error handling and logging for better troubleshooting
Fixed JSON import issues and other technical bugs

This approach allows Bedrock Agents to access files of any size through their /get-file endpoint, making the system more efficient and robust.

2. Bedrock Agent Management Functionality

We've added new capabilities to discover and interact with Bedrock Agents:

Agent discovery: When Bedrock Agent is enabled in the config without specific agent parameters, the system lists all available agents in the account
Targeted agent selection: When enabled with a specific agent ID, the system lists only that agent
Multi-function support: The system can now work with multiple Bedrock Agents, allowing different agents to be used for different purposes

These enhancements make it easier to discover, select, and use Bedrock Agents within the chatbot framework.

Technical Implementation Details

Modified the BedrockAgentAdapter class to handle S3 URIs instead of file content
Enhanced the bedrock_agent/client.py module with agent discovery and selection capabilities
Added proper error handling for non-English text and other edge cases
Implemented comprehensive logging for debugging and monitoring
Ensured backward compatibility with existing configurations

This pull request significantly improves the chatbot's ability to work with Bedrock Agents and handle file-based interactions more efficiently.

maryamkhidir

Looks good!

Please add unit tests at least for new logic and we should also have some integration tests for the Agentic flow.
And fix linting errors - npm run vet-all run all the checks.

maryamkhidir · 2025-08-22T08:52:51Z

lib/model-interfaces/langchain/functions/request-handler/adapters/bedrock_agent/agent.py

Please add test coverage for this file

maryamkhidir · 2025-08-22T08:54:03Z

lib/shared/layers/python-sdk/python/genai_core/bedrock_agent/client.py

Please add test coverage for this

yosefkri and others added 2 commits August 11, 2025 10:21

Ft/bedrock agent (#1)

29083e9

* Adding support using bedrock agent * update langchain issue * updated plan readme --------- Co-authored-by: Yossi Kricheli <yossik@amazon.com>

Merge branch 'aws-samples:main' into main

6ad9792

maryamkhidir reviewed Aug 12, 2025

View reviewed changes

maryamkhidir self-requested a review August 12, 2025 08:18

maryamkhidir reviewed Aug 12, 2025

View reviewed changes

Ft/bedrock agent (#2)

498a15c

* Adding support using bedrock agent * update langchain issue * updated plan readme * Making code more readable --------- Co-authored-by: Yossi Kricheli <yossik@amazon.com>

Rob-Powell reviewed Aug 12, 2025

View reviewed changes

yosefkri and others added 2 commits August 21, 2025 14:51

Merge branch 'main' into main

d349e8e

maryamkhidir requested changes Aug 22, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add Amazon Bedrock Agent Support #676

Add Amazon Bedrock Agent Support #676

Uh oh!

yosefkri commented Aug 12, 2025

Uh oh!

maryamkhidir Aug 12, 2025

Uh oh!

maryamkhidir Aug 12, 2025

Uh oh!

maryamkhidir Aug 12, 2025

Uh oh!

maryamkhidir left a comment •

edited

Loading

Uh oh!

Rob-Powell left a comment

Uh oh!

yosefkri commented Aug 13, 2025

Uh oh!

yosefkri commented Aug 21, 2025

Uh oh!

maryamkhidir left a comment •

edited

Loading

Uh oh!

maryamkhidir Aug 22, 2025

Uh oh!

maryamkhidir Aug 22, 2025

Uh oh!

Uh oh!

Add Amazon Bedrock Agent Support #676

Are you sure you want to change the base?

Add Amazon Bedrock Agent Support #676

Uh oh!

Conversation

yosefkri commented Aug 12, 2025

Overview

Key Changes

Testing

Challenges Addressed

Documentation

How to Test

Future Enhancements

Uh oh!

maryamkhidir Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

maryamkhidir Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

maryamkhidir Aug 12, 2025

Choose a reason for hiding this comment

Uh oh!

maryamkhidir left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Rob-Powell left a comment

Choose a reason for hiding this comment

Uh oh!

yosefkri commented Aug 13, 2025

Uh oh!

yosefkri commented Aug 21, 2025

Pull Request Summary: Enhanced Bedrock Agent Support

1. S3 URI-Based File Handling for Bedrock Agents

2. Bedrock Agent Management Functionality

Technical Implementation Details

Uh oh!

maryamkhidir left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

maryamkhidir Aug 22, 2025

Choose a reason for hiding this comment

Uh oh!

maryamkhidir Aug 22, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

maryamkhidir left a comment •

edited

Loading

maryamkhidir left a comment •

edited

Loading