hunfabric/modules/agent-engine

Agent Engine Module

The module creates Agent Engine and related dependencies.

• It supports both source based deployments (aka in-line deployment) and serialized object deployment (aka pickle deployment).
• For serialized object deployment, the module creates a GCS bucket to store the pickled object and related dependencies.
• It supports Customer Managed Encryption Keys (CMEK) to encrypt both the reasoning engine and the GCS bucket.
• It provides support for both managed and unmanaged (Terraform doesn't track updates to code) deployments.
• It provides support for VPC-SC (via PSC-I).
• It provides support for custom and default service accounts.
• It provides support for environment variables and secrets from Secret Manager.
• It supports both Python-based and container-based deployments.

TOC

• TOC
• Minimal deployment
• Serialized Object Deployment
• Unmanaged deployments
• Service accounts
• Private networking: setup PSC-I
• Specify an encryption key
• Define environment variables and use secrets
• Container-based deployment
• Memory Bank
• Getting values from context
• Disable deletion protection
• Variables
• Outputs

Minimal deployment

This example shows how to deploy an agent engine with minimal configuration, using source code from a local path.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }
}
# tftest inventory=minimal.yaml

You can change the name of the tar.gz package, of the requirement file, the name of the Python file and the name of the agent function by using the deployment_config.source_files_config variable.

You can also provide custom build arguments for the container image by using the deployment_config.source_files_config.image_spec variable.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
      image_spec = {
        build_args = {
          "ENV" = "production"
        }
      }
    }
  }
}
# tftest inventory=image-spec.yaml

Serialized Object Deployment

You can also manually serialize your agent by using the cloudpickle library and pass the pickle.pkl, dependencies.tar.gz and requirements.txt files to the module.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    package_config = {
      pickle_path       = "assets/src/pickle.pkl"
      dependencies_path = "assets/src/dependencies.tar.gz"
      requirements_path = "assets/src/requirements.txt"
    }
  }
}
# tftest inventory=minimal-pickle.yaml

If the files are already in a GCS bucket, you can pass the GCS URIs to the module.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    package_config = {
      are_paths_local   = false
      pickle_path       = "gs://my-bucket/pickle.pkl"
      dependencies_path = "gs://my-bucket/dependencies.tar.gz"
      requirements_path = "gs://my-bucket/requirements.txt"
    }
  }
}
# tftest inventory=pickle-gcs.yaml

Unmanaged deployments

If you want to use the module just to bootstrap the infrastructure and then manage the code updates yourself, you can set the managed variable to false.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region
  managed    = false

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }
}
# tftest inventory=unmanaged.yaml

Service accounts

You can choose to use a custom service account or let the module create one for you.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }
}
# tftest inventory=sa-default.yaml

Using a custom service account.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }

  service_account_config = {
    create = false
    email  = "my-agent@project-id.iam.gserviceaccount.com"
  }
}
# tftest inventory=sa-custom.yaml

Private networking: setup PSC-I

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }

  networking_config = {
    network_attachment_id = "projects/project-id/regions/europe-west8/networkAttachments/my-nat"
    dns_peering_configs = {
      "googleapis.com." = {
        target_network_name = "my-network"
      }
    }
  }
}
# tftest inventory=psc-i.yaml

Specify an encryption key

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }

  encryption_key = "projects/project-id/locations/europe-west8/keyRings/my-keyring/cryptoKeys/my-key"
}
# tftest inventory=encryption.yaml

Define environment variables and use secrets

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
    environment_variables = {
      FOO = "bar"
    }
    secret_environment_variables = {
      MY_SECRET = {
        secret_id = "projects/project-id/secrets/my-secret"
      }
    }
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }
}
# tftest inventory=environment.yaml

Container-based deployment

You can deploy your agent as a custom Docker image.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    environment_variables = {
      FOO = "bar"
    }
  }

  deployment_config = {
    container_config = {
      image_uri = "us-central1-docker.pkg.dev/my-project/my-repo/my-image:latest"
    }
  }
}
# tftest inventory=container.yaml

Memory Bank

You can optionally configure a Memory Bank to provide long-term persistent memory for your agent.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = var.project_id
  region     = var.region

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }

  memory_bank_config = {
    disable_memory_revisions = false
    generation_config = {
      model = "projects/my-project/locations/us-central1/publishers/google/models/gemini-2.0-flash-001"
    }
    similarity_search_config = {
      embedding_model = "projects/my-project/locations/us-central1/publishers/google/models/text-embedding-005"
    }
    ttl_config = {
      default_ttl = "2592000s" # 30 days
    }
  }
}

Getting values from context

The module allows you to dynamically reference context values for resources created outside this module, through the context variable. This includes the definition of custom roles, iam_principals, locations, networks, psc_network_attachments, kms_keys, models and project ids.

module "agent_engine" {
  source     = "./fabric/modules/agent-engine"
  name       = "my-agent"
  project_id = "$project_ids:main-project"
  region     = "$locations:primary"
  agent_engine_config = {
    agent_framework = "google-adk"
  }
  deployment_config = {
    source_files_config = {
      source_path = "assets/src/source.tar.gz"
    }
  }
  networking_config = {
    network_attachment_id = "$psc_network_attachments:primary"
    dns_peering_configs = {
      "example.com" = {
        target_network_name = "$networks:vpc-1"
      }
      "my-company.local" = {
        target_network_name = "$networks:vpc-2"
        target_project_id   = "$project_ids:dns-project"
      }
    }
  }
  service_account_config = {
    create = false
    email  = "$iam_principals:my-custom-sa"
  }
  context = {
    iam_principals = {
      my-custom-sa = "my-sa@$test-project-1.iam.gserviceaccount.com"
    }
    locations = {
      primary = "europe-west1"
    }
    networks = {
      vpc-1 = "my-vpc-1"
      vpc-2 = "my-vpc-2"
    }
    project_ids = {
      main-project = "test-project-1"
      dns-project  = "company-dns-project"
    }
    psc_network_attachments = {
      primary = "projects/test-project-1/regions/europe-west1/networkAttachments/core-service"
    }
  }
}
# tftest inventory=context.yaml

Disable deletion protection

By default you can't neither delete your agent if it has session or your GCS bucket if it has files inside. For testing, you can anyway force the deletion of these resources:

module "agent_engine" {
  source                     = "./fabric/modules/agent-engine"
  name                       = "my-agent"
  project_id                 = var.project_id
  region                     = var.region
  enable_deletion_protection = false

  agent_engine_config = {
    agent_framework = "google-adk"
  }

  deployment_config = {
    package_config = {
      pickle_path       = "assets/src/pickle.pkl"
      dependencies_path = "assets/src/dependencies.tar.gz"
      requirements_path = "assets/src/requirements.txt"
    }
  }
}
# tftest inventory=deletion-protection.yaml

Variables

name	description	type	required	default
name	The name of the agent.	string	✓
project_id	The id of the project where to deploy the agent.	string	✓
region	The region where to deploy the agent.	string	✓
agent_engine_config	The agent configuration. Supported values for agent_framework: 'google-adk', 'langchain', 'langgraph', 'ag2', 'llama-index', 'custom'.	object({…})		{}
bucket_config	The GCS bucket configuration.	object({…})		{}
context	Context-specific interpolations.	object({…})		{}
deployment_config	The deployment configuration.	object({…})		{}
description	The Agent Engine description.	string		"Terraform managed."
enable_deletion_protection	Whether deletion protection should be enabled.	bool		true
encryption_key	The full resource name of the Cloud KMS CryptoKey.	string		null
managed	Whether the Terraform module should control the code updates.	bool		true
memory_bank_config	Configuration for the memory bank.	object({…})		null
networking_config	Networking configuration.	object({…})		null
service_account_config	Service account configurations.	object({…})		{}

Outputs

name	description	sensitive
agent	The Agent Engine object.
id	Fully qualified Agent Engine id.
service_account	Service account resource.