Can't 'store' data into Knowledge base.

[trey-rosius]

Hi folks, I have a bedrock knowledge base connected to a pinecone vector store and an S3 datasource. I'm able to add documents through the AWS Console and query the KB.
But using Strands, i can't add data to the KB.
This is what i get

No Logs or proper feedback on what's actually going on.

Here's the code

import asyncio
import json
import os

from strands import Agent
from strands_tools import http_request
from strands_tools import use_llm, memory
from strands.models import BedrockModel

SYSTEM_PROMPT = """

"""


bedrock_model = BedrockModel(
    model_id="anthropic.claude-3-5-sonnet-20240620-v1:0",
    region_name='us-east-1',
    temperature=0.3,
    cache_tools='default',

)


# Create an agent with HTTP capabilities
weather_agent = Agent(
    model=bedrock_model,
    system_prompt=SYSTEM_PROMPT,
    callback_handler=None,
    tools=[use_llm, memory],  # Explicitly enable http_request tool
)

weather_agent.tool.memory(
        action="store",
        STRANDS_KNOWLEDGE_BASE_ID="U1REFHCZG2",
        AWS_REGION="us-east-1",
        content="loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there's uh there's SQL and there's no SQL. The SQL database is like MySQL. And then there's a new SQL database like ",
        title="Meeting Notes",

    )

Can i please get some help ?

[awsarron]

https://github.com/strands-agents/tools/blob/af3314e9997d2ac884ed4983e3b953446c2fcb7a/src/strands_tools/memory.py#L564 the parameter for region name is region_name instead of AWS_REGION. Does changing that help?

weather_agent.tool.memory(
    action="store",
    STRANDS_KNOWLEDGE_BASE_ID="U1REFHCZG2",
    region_name="us-east-1",
    content="loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there's uh there's SQL and there's no SQL. The SQL database is like MySQL. And then there's a new SQL database like ",
    title="Meeting Notes",
)

Enabling debug logs can show more verbose output: https://strandsagents.com/latest/user-guide/quickstart/#debug-logs

[trey-rosius]

@awsarron nothing changes. Here's the output i get

(.venv) ➜  strands-intro python -u agent_kb.py
DEBUG | strands.tools.registry | getting tool configurations
DEBUG | strands.tools.registry | tool_name=<memory> | loaded tool config
DEBUG | strands.tools.registry | tool_name=<use_llm> | loaded tool config
DEBUG | strands.tools.registry | tool_count=<2> | tools configured
DEBUG | strands.handlers.tool_handler | tool=<{'toolUseId': 'tooluse_memory_860113641', 'name': 'memory', 'input': {'action': 'store', 'STRANDS_KNOWLEDGE_BASE_ID': 'U1REFHCZG2', 'region_name': 'us-east-1', 'content': "loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there's uh there's SQL and there's no SQL. The SQL database is like MySQL. And then there's a new SQL database like Dynamoibi, so it's super fast and it simply just helps you save and retrieve your data for your applications with ease. A simple scenario is always, you know, creating a rest API that saves, that creates. Reads updates. Deletes and lists. Items from a Dynamo DB table. And another scenario is you can actually save images into an S3 bucket and then save the link to those images in a Dynamo DB table and use the Lambda function now to retrieve them. If you really want to know a lot more about Dynamo DB tables, lambda functions, estrays, and all that, you can go to the official documentation where you can watch maybe other workshops on Erolo Academy to see how we use them practically", 'title': 'Meeting Notes'}}> | invoking
Do you want to proceed with the store operation? [y/*] y
DEBUG | strands.agent.conversation_manager.sliding_window_conversation_manager | window_size=<4>, message_count=<40> | skipping context reduction

[awsarron]

Ah that's unfortunate. What is returned by the agent.tool.memory() call?

[trey-rosius]

Nothing. It's like a blackbox. No logs.

[awsarron]

Sorry for the trouble, we'll get to the bottom of it and improve logging to make it easier to debug in the future.

Can you share the output of this snippet please:

import asyncio
import json
import os

from strands import Agent
from strands_tools import http_request
from strands_tools import use_llm, memory
from strands.models import BedrockModel

SYSTEM_PROMPT = """

"""


bedrock_model = BedrockModel(
    model_id="anthropic.claude-3-5-sonnet-20240620-v1:0",
    region_name='us-east-1',
    temperature=0.3,
    cache_tools='default',

)


# Create an agent with HTTP capabilities
weather_agent = Agent(
    model=bedrock_model,
    system_prompt=SYSTEM_PROMPT,
    callback_handler=None,
    tools=[use_llm, memory],  # Explicitly enable http_request tool
)

tool_result = weather_agent.tool.memory(
    action="store",
    STRANDS_KNOWLEDGE_BASE_ID="U1REFHCZG2",
    region_name="us-east-1",
    content="loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there's uh there's SQL and there's no SQL. The SQL database is like MySQL. And then there's a new SQL database like ",
    title="Meeting Notes",
)

print("\n\n")
print("TOOL RESULT:")
print(tool_result)
print("\n\n")
print("AGENT MESSAGES:")
print(agent.messages)

[trey-rosius]

(.venv) ➜  strands-intro python -u snippet.py
Do you want to proceed with the store operation? [y/*] y



TOOL RESULT:
{'status': 'error', 'content': [{'text': "❌ Error during store operation: An error occurred (ValidationException) when calling the IngestKnowledgeBaseDocuments operation: The dataSourceType that you specified in CUSTOM doesn't match the type of the data source specified in the request header. Check that both types match and retry your request."}], 'toolUseId': 'tooluse_memory_884194805'}



AGENT MESSAGES:
[{'role': 'user', 'content': [{'text': 'agent.tool.memory direct tool call.\nInput parameters: {"action": "store", "STRANDS_KNOWLEDGE_BASE_ID": "U1REFHCZG2", "region_name": "us-east-1", "content": "loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there\'s uh there\'s SQL and there\'s no SQL. The SQL database is like MySQL. And then there\'s a new SQL database like ", "title": "Meeting Notes"}\n'}]}, {'role': 'assistant', 'content': [{'toolUse': {'toolUseId': 'tooluse_memory_884194805', 'name': 'memory', 'input': {'action': 'store', 'STRANDS_KNOWLEDGE_BASE_ID': 'U1REFHCZG2', 'region_name': 'us-east-1', 'content': "loaded data Dynamo DB is a super fast, no SQL, no SQL database. No, there's uh there's SQL and there's no SQL. The SQL database is like MySQL. And then there's a new SQL database like ", 'title': 'Meeting Notes'}}}]}, {'role': 'user', 'content': [{'toolResult': {'status': 'error', 'content': [{'text': "❌ Error during store operation: An error occurred (ValidationException) when calling the IngestKnowledgeBaseDocuments operation: The dataSourceType that you specified in CUSTOM doesn't match the type of the data source specified in the request header. Check that both types match and retry your request."}], 'toolUseId': 'tooluse_memory_884194805'}}]}, {'role': 'assistant', 'content': [{'text': 'agent.memory was called'}]}]

[JackYPCOnline]

Thank you for providing the detailed feedback and logs; they were instrumental in diagnosing the issue.

My investigation confirms your analysis is correct. The error CUSTOM doesn't match the type of the data source specified occurs because the memory tool currently hardcodes the dataSourceType to CUSTOM when performing the store action. This creates a mismatch, as your Knowledge Base (U1REFHCZG2) is correctly configured to use the S3 data source type.

This is a limitation in our tool, not an error in your setup.

We have logged this as a bug and are prioritizing a fix. The correct long-term solution is for the tool to be updated to intelligently detect the data source type from the Knowledge Base configuration before sending the request.

[trey-rosius]

Thank you @JackYPCOnline . I'll be waiting on the update then.

[trey-rosius]

Hi @JackYPCOnline , is there an ETA for this fix ? Sorry if i'm being impatient. I understand there's a lot of work to be done.

The thing is, i'm working on a huge course/workshop for Strands and i wanted to showcase both sides of it's RAG feature(store and retrieve).

Of course i can use other means to save the data into the KB. But i particularly wanted to showcase how it's done with Strands.

[JackYPCOnline]

Hi @trey-rosius thank you for your patience while we looked into this. I have an update for you after a deeper discussion with our engineering team.

After reviewing the architecture, the team has clarified that the memory tool's current behavior—hardcoding the data source to CUSTOM—is the intended design. Therefore, we will not be changing it to support S3 data sources directly from the store action.
The fundamental reason for this decision is that CUSTOM and S3 data sources are designed for different purposes and workflows:

• The S3 data source is designed for managing larger, file-based knowledge. The process involves two steps: first uploading the file(s) to your S3 bucket, and then running a sync operation to index that data into the Knowledge Base.
• The CUSTOM data source is specifically designed for agents to use on-the-fly. It allows an agent to instantly "remember" or store small pieces of information using a direct push command, without a separate sync step.
  The memory tool is optimized for this second, real-time agent workflow. Supporting S3 would introduce the complexity of managing uploads and syncs, which is counter to the tool's purpose of simple, immediate data storage.

To showcase the store action specifically, you could configure a separate Knowledge Base that uses the CUSTOM data source type and use it to demonstrate how an agent can learn and save new facts in real-time.

We apologize for the initial miscommunication where this was labeled a bug. It was a misunderstanding of the tool's intended design.

[trey-rosius]

@JackYPCOnline thanks for the clarification. I'll go with Custom Datasources then.