terraform-aws-arc-bedrock ¶

Module: sourcefuse/arc-bedrock/aws

Registry: https://registry.terraform.io/modules/sourcefuse/arc-bedrock/aws

Category: AI / Machine Learning

Source: https://github.com/sourcefuse/terraform-aws-arc-bedrock

Overview¶

This Terraform module provides an automated way to deploy and manage AWS Bedrock Agents, Collaborators, Action Groups, and associated IAM Roles. The module is designed to be flexible, reusable, and configurable to suit various use cases.

Features¶

Bedrock Agent Creation: Deploys an AWS Bedrock agent with configurable settings.
Collaborator Management: Supports multiple collaborators with required IAM permissions.
Action Groups: Enables defining multiple action groups for executing Lambda-based functions.
IAM Role Management: Automatically provisions IAM roles with necessary policies for Bedrock resources.

What It Does¶

Bedrock Agent with configurable foundation model and instructions
Multi-agent collaboration (supervisor + collaborator pattern)
Action groups with Lambda function schemas
Knowledge base with S3 data sources and OpenSearch Serverless vector store
IAM roles auto-provisioned for agents

For more information about this repository and its usage, please see Terraform AWS BEDROCK Usage Guide.

Quickstart¶

Module Components¶

1. Bedrock Agent¶

Configurable name, foundation model, and instruction set.
Supports enabling/disabling agent collaboration.
Customizable session timeout settings.

2. Collaborators¶

Allows defining multiple collaborators with their own permissions.
IAM roles and policies created dynamically.

3. Action Groups¶

Enables defining multiple action groups.
Supports dynamic function schemas for action execution.
Integrates AWS Lambda functions.

4. IAM Roles & Policies¶

Automatically provisions IAM roles for agents and collaborators.
Grants necessary permissions for Bedrock services.

Usage¶

Prerequisites¶

Before using this module, ensure you have the following:

AWS credentials configured.
Terraform installed.
A working knowledge of Terraform.

Getting Started¶

Define the Module

Initially, it's essential to define a Terraform module, which is organized as a distinct directory encompassing Terraform configuration files. Within this module directory, input variables and output values must be defined in the variables.tf and outputs.tf files, respectively. The following illustrates an example directory structure:

bedrock-agent/
|-- main.tf
|-- variables.tf
|-- outputs.tf
|-- locals.tf

Define Input Variables

Inside the variables.tf or in *.tfvars file, you should define values for the variables that the module requires.

Use the Module in Your Main Configuration In your main Terraform configuration file (e.g., main.tf), you can use the module. Specify the source of the module, and version, For Example

module "bedrock_agents" {
  source = "sourcefuse/arc-bedrock/aws"
  version = "0.0.1"

  bedrock_agent_config = {
    create              = true
    name                = "arc-bedrock-agent"
    foundation_model    = "anthropic.claude-3-5-sonnet-20241022-v2:0"
    instruction         = "You are a customer support assistant. Answer user queries."
    agent_collaboration = "SUPERVISOR"
    prepare_agent       = false
    description         = "Supervisor agent"
  }
  agent_collaborator = {
    name                        = "collab-1"
    collaborator_name           = "Collaborator-One"
    foundation_model            = "anthropic.claude-3-5-sonnet-20241022-v2:0"
    instruction                 = "do what the supervisor is asking you to do"
    collaboration_instruction   = "tell the other agent on what to do"
    alias_name                  = "DocProcessor"
    description                 = "Collaborator 1"
    relay_conversation_history  = "TO_COLLABORATOR"
    prepare_agent               = true
    idle_session_ttl_in_seconds = 600
    action_groups               = local.action_groups
  }

  tags = module.tags.tags
}

Required Inputs¶

Name	Type	Description
`namespace`	`string`	Namespace prefix
`environment`	`string`	Deployment environment
## Key Outputs

Name	Description
`agent_id`	Bedrock Agent ID
`agent_arn`	Bedrock Agent ARN
`agent_role_arn`	IAM role ARN for the agent
## Full Variable & Output Reference

The complete inputs/outputs reference is auto-generated below.

Output Values

Inside the outputs.tf file of the module, you can define output values that can be referenced in the main configuration. For example:

output "agent_arn" {
  value       = module.bedrock_agent.agent_arn
  description = "Agent arn"
}

output "agent_id" {
  value       = module.bedrock_agent.agent_id
  description = "Agent ID"
}

output "collaborator_agent_id" {
  description = "Agent ID created for collaborators."
  value       = module.bedrock_agent.collaborator_agent_id
}

output "agent_role_arn" {
  value       = module.bedrock_agent.agent_role_arn
  description = "Agent Role arn"
}

.tfvars

Inside the .tfvars file of the module, you can provide desired values that can be referenced in the main configuration. For example:

Edit the locals.tf file and provide desired values.

collaborators - List of Collaborator Agents

action_groups - Action groups for Collaborator Agents

locals {

  collaborators = [
    {
      name                        = "collab-2"
      supervisor_agent_id         = module.bedrock_agent.agent_id
      collaborator_name           = "Collaborator-Two"
      foundation_model            = "anthropic.claude-3-5-sonnet-20241022-v2:0"
      instruction                 = "do what the supervisor is asking you to do"
      collaboration_instruction   = "tell the other agent on what to do"
      alias_name                  = "TechSupport"
      description                 = "Collaborator 2"
      relay_conversation_history  = "TO_COLLABORATOR"
      prepare_agent               = true
      idle_session_ttl_in_seconds = 600
      action_groups               = local.action_groups
    },
    {
      name                        = "collab-3"
      supervisor_agent_id         = module.bedrock_agent.agent_id
      collaborator_name           = "Collaborator-Three"
      foundation_model            = "anthropic.claude-3-5-sonnet-20241022-v2:0"
      instruction                 = "do what the supervisor is asking you to do"
      collaboration_instruction   = "tell the other agent on what to do"
      alias_name                  = "TechSupport"
      description                 = "Collaborator 3"
      relay_conversation_history  = "TO_COLLABORATOR"
      prepare_agent               = true
      idle_session_ttl_in_seconds = 600
      action_groups               = local.action_groups
    }
  ]

  action_groups = [{
    name                       = "singlerulegenerationagent-actiongroup"
    state                      = "ENABLED"
    agent_version              = "DRAFT"
    skip_resource_in_use_check = true
    action_group_executor      = { lambda = "arn:aws:lambda:us-east-1:884360309640:function:arc-debug-budgets-default" }

    function_schema = [
      {
        functions = [
          {
            name        = "extract_general_props"
            description = "Extracts general info properties based on the user prompt."
            parameters = [
              {
                map_block_key = "prompt"
                type          = "string"
                description   = "The user instruction/prompt to create or edit the rule."
                required      = true
              },
              {
                map_block_key = "rule_context"
                type          = "string"
                description   = "The existing rule to be updated, if provided."
                required      = false
              }
            ]
          },
          {
            name        = "extract_static_props"
            description = "Extracts static properties based on the user prompt."
            parameters = [
              {
                map_block_key = "prompt"
                type          = "string"
                description   = "The user instruction/prompt to create or edit the rule."
                required      = true
              },
              {
                map_block_key = "rule_context"
                type          = "string"
                description   = "The existing rule to be updated, if provided."
                required      = false
              }
            ]
          }
        ]
      }
    ]
  }]

}

First Time Usage¶

uncomment the backend block in main.tf

terraform init -backend-config=config.dev.hcl

If testing locally, terraform init should be fine

Create a dev workspace

terraform workspace new dev

Plan Terraform

terraform plan -var-file dev.tfvars

Apply Terraform

terraform apply -var-file dev.tfvars

Production Setup¶

terraform init -backend-config=config.prod.hcl

Create a prod workspace

terraform workspace new prod

Plan Terraform

terraform plan -var-file prod.tfvars

Apply Terraform

terraform apply -var-file prod.tfvars  -parallelism=1

Cleanup¶

Destroy Terraform

terraform destroy -var-file dev.tfvars -parallelism=1

Module Usage¶

To see a full example, check out the main.tf file in the example folder.

Requirements¶

Name	Version
terraform	>= 1.3, < 2.0.0
aws	>= 5.0, < 6.0
opensearch	2.3.1
time	0.13.0

Providers¶

Name	Version
aws	5.89.0

Modules¶

Name	Source	Version
collaborators	./modules/collaborator	n/a
knowledge_base	./modules/knowledge-base	n/a

Resources¶

Name	Type
aws_bedrockagent_agent.collaborator	resource
aws_bedrockagent_agent.this	resource
aws_bedrockagent_agent_alias.this	resource
aws_iam_role.collaborator	resource
aws_iam_role.this	resource
aws_iam_role_policy.collaborator	resource
aws_iam_role_policy.this	resource
aws_caller_identity.current	data source
aws_iam_policy_document.agent_permissions	data source
aws_iam_policy_document.agent_trust	data source
aws_iam_policy_document.collaborator_agent_permissions	data source
aws_partition.current	data source
aws_region.current	data source

Inputs¶

Name	Description	Type	Default	Required
agent_collaborator	Configuration object for a collaborator, including name, instructions, and settings.	object({ name = string supervisor_agent_id = optional(string, null) collaborator_name = optional(string, null) instruction = string collaboration_instruction = string alias_name = string foundation_model = string description = optional(string, "") relay_conversation_history = optional(string, "TO_COLLABORATOR") prepare_agent = optional(bool, true) idle_session_ttl_in_seconds = optional(number, 500) action_groups = optional(list(object({ name = string state = string agent_version = string skip_resource_in_use_check = optional(bool, true) action_group_executor = object( { lambda = optional(object({ name = string add_permission = optional(bool, true) })) custom_control = optional(string, null) }) function_schema = list(object({ functions = list(object({ name = string description = string parameters = list(object({ map_block_key = string type = string description = string required = bool })) })) })) })), []) })	`null`	no
bedrock_agent_config	Configuration for the Amazon Bedrock Agent, including name, session TTL, foundation model, tags, instructions, collaboration settings, and preparation options.	object({ create = optional(bool, false) name = optional(string, null) alias_name = optional(string, null) alias_description = optional(string, null) idle_session_ttl_in_seconds = optional(number, 500) foundation_model = optional(string, null) instruction = optional(string, null) agent_collaboration = optional(string, "DISABLED") description = optional(string, null) prepare_agent = optional(bool, true) role_arn = optional(string, null) })	{ "create": false }	no
environment	Name of the environment, i.e. dev, stage, prod	`string`	n/a	yes
knowledge_base_config	Configuration for AWS Bedrock Agent Knowledge Base, including vector storage, embedding model, and OpenSearch integration.	object({ create = optional(bool, false) name = string role_arn = optional(string, null) agent_role_name = optional(string, null) foundation_model_arn = string description = optional(string, null) agent_id = optional(string, null) instruction = string data_source_list = list(object({ type = optional(string, "S3") s3_config = optional(object({ create = optional(bool, false) name = string inclusion_prefixes = optional(list(string), []) })) })) data_storage_list = optional(list(object({ type = optional(string, "S3") s3_config = optional(object({ create = optional(bool, false) prefix = optional(string, "") name = string })) })), []) embedding_model_configuration = object({ dimensions = optional(number, 1024) embedding_data_type = string }) storage_configuration = object({ type = optional(string, "OPENSEARCH_SERVERLESS") opensearch_serverless_configuration = object({ create = optional(bool, false) name = optional(string, null) collection_arn = optional(string, null) access_policy_rules = optional(list(any), []) data_lifecycle_policy_rules = optional(list(any), []) index_config = object({ number_of_shards = optional(string, "2") number_of_replicas = optional(string, "0") index_knn = optional(bool, true) index_knn_algo_param_ef_search = optional(string, "512") mappings = optional(string, null) }) vector_index_name = string field_mapping = object({ vector_field = string text_field = string metadata_field = string }) }) }) })	{ "create": false, "data_source_list": [], "data_storage_list": [], "embedding_model_configuration": null, "foundation_model_arn": null, "instruction": null, "name": null, "storage_configuration": null }	no
namespace	Namespace of the project, i.e. arc	`string`	n/a	yes
tags	Tags for Bedrock resources	`map(string)`	`{}`	no

Outputs¶

Name	Description
agent_arn	Agent arn
agent_id	Agent ID
agent_role_arn	Agent Role arn
alias_arn	ARN of the alias
alias_id	Unique identifier of the alias.
collaborator_agent_id	Agent ID created for collaborators.
collaborator_role_arns	ARNs of the IAM roles created for collaborators.
opensearch_collection_endpoint	Opensearch Collection endpoint

Development¶

Prerequisites¶

Configurations¶

Configure pre-commit hooks

1	`pre-commit install`

Git commits¶

while Contributing or doing git commit please specify the breaking change in your commit message whether its major,minor or patch

For Example

git commit -m "your commit message #major"

By specifying this , it will bump the version and if you dont specify this in your commit message then by default it will consider patch and will bump that accordingly

Tests¶

Tests are available in test directory
Configure the dependencies

cd test
go mod init github.com/sourcefuse/terraform-aws-arc-bedrock
go get github.com/gruntwork-io/terratest/modules/terraform

Now execute the test

cd test/
go test

Contributing¶

See CONTRIBUTING.md for commit conventions and development setup.

Authors¶

This project is authored by: - SourceFuse ARC Team

terraform-aws-arc-bedrock¶

Overview¶

Features¶

What It Does¶

Quickstart¶

Module Components¶

1. Bedrock Agent¶

2. Collaborators¶

3. Action Groups¶

4. IAM Roles & Policies¶

Usage¶

Prerequisites¶

Getting Started¶

Required Inputs¶

First Time Usage¶

Production Setup¶

Cleanup¶

Module Usage¶

Requirements¶

Providers¶

Modules¶

Resources¶

Inputs¶

Outputs¶

Development¶

Prerequisites¶

Configurations¶

Git commits¶

Tests¶

Contributing¶

Authors¶

terraform-aws-arc-bedrock ¶