Create agents that communicate to each other and have long term memory

• Communication replies on Valkey pub/sub
• Persistent short memory storage uses Apache Kafka topics
• Long term memory relies on RAG, langchain, OpenSearch and Apache Kafka

Step 1: Create a workspace with GitPod

https://gitpod.io/new/#https://github.com/Aiven-Labs/agents-with-memories-rag-valkey-opensearch-bedrock

Step 2: Deploy Aiven services with Terraform

Register with Aiven

If you don't have Aiven account yet, register here to get extra credits.

Get Aiven token

In Aiven console Go to User Information menu and select Tokens:

Click on Generate token, add a description for the token and press to generate. A popup with a newly generated token will appear. Copy token value.

Run Terraform script

In this project we use Terraform to set up Aiven services. If you don't have Terraform installed follow the steps from the Terraform documentation page.

The terraform files can be found in ./terraform folder of the current project. ./terraform/terraform.tfvars-example provides an example of tfvars file.

1. Rename (or create a new file) to ./terraform/terraform.tfvars
2. Populate aiven_api_token with the token that you copied from the console
3. Populate project_name with the name of the Aiven project
4. Navigate to terraform folder
5. Set export PROVIDER_AIVEN_ENABLE_BETA=true in the terminal (Terraform Valkey resource is still in beta stage)
6. Run terraform init
7. Run terraform plan
8. Run terraform apply

Terraform will initiate creation of four resources:

• Aiven for Apache Kafka
• Apache Kafka topic
• Aiven for OpenSearch
• Aiven for caching (Valkey)

Once deployment is done, Terraform will also create .env file with necessary credentials to access the services and certificates folder with Apache Kafka certificates.

Step 3: Enable Amazon Bedrock model Claude

In this project we use LLM Claude available through Amazon Bedrock. To invoke the model you need first to enable access:

1. In AWS console select region where Amazon Bedrock and Claude model is available (for example, us-east-1).
2. Go to Amazon Bedrock service page.
3. Select Providers in the menu and then Anthropic, scroll to the Claude models.
4. In this project we use Claude 3 Haiku, you can also select a different model, but that might require changes in request/response formats.
5. If you see a message This account does not currently have access to this model. Request access in Model access., go to Model access and enable the model.

AWS credentials

This project assumes that AWS_SECRET_ACCESS_KEY and AWS_ACCESS_KEY_ID are accessible through the environment variables. Follow AWS documentation for more details.

Context: the credentials are used by @aws-sdk/client-bedrock-runtime when creating a client, see Bedrock client creation in ./src/common.js for details.

Step 4: Install libraries

This project uses npm and NodeJS, if you don't have them installed, follow instruction at the NodeJS documentation page Run npm install

Step 5: Run locally

Run node run