How to update configuration (env vars) / prompts of an already deployed Agent Engine resource

[Siddharthan-K]

I am attempting to modify the things of an already deployed Agent Engine resource, specifically to update environment variables, without redeploying the agent.

Using the Python SDK, calling agent_engines.update() with only partial config data results in a FAILED_PRECONDITION error (package_spec cannot be updated to empty). Fetching the existing agent and merging the current config before updating still does not work reliably, as the config schema is not clearly exposed.

• Is updating an existing Agent Engine supported?
• If yes, what is the correct and supported way to preserve package_spec and other required fields during an update?
• If not supported, is redeployment the only intended approach for updates?

[adk-bot]

Response from ADK Answering Agent (experimental, answer may be inaccurate)

TLDR: Redeploying your agent using adk deploy agent_engine is the recommended and supported way to update environment variables and other configurations. You should not call the agent_engines.update() method directly.

---

Hello! I see you're having trouble updating your Agent Engine resource. Based on the ADK documentation and examples, here's the recommended approach:

To update the environment variables or any other configuration of an already deployed Agent Engine, you should redeploy your agent using the adk deploy agent_engine command. The ADK CLI handles the complexity of packaging your agent's code and configuration, including the package_spec, which is a required field for updates. [2]

Here are the steps to follow:

1. Update your local configuration: Modify the .env file in your agent's project directory with the new environment variable values. [3, 6]

2. Redeploy the agent: Run the adk deploy agent_engine command again, but this time, include the --agent-engine-id flag with the ID of your existing Agent Engine instance. The command will look something like this:

   adk deploy agent_engine \
     --project=$GOOGLE_CLOUD_PROJECT \
     --region=$GOOGLE_CLOUD_LOCATION \
     --staging_bucket=$STAGING_BUCKET \
     --agent-engine-id=<YOUR_AGENT_ENGINE_ID> \
     $AGENT_PATH

When you provide the agent_engine_id, the ADK CLI will update the existing deployment with your new configuration instead of creating a new one. [2]

Calling agent_engines.update() directly with partial data is not the intended workflow, which is why you are encountering the FAILED_PRECONDITION error. The adk deploy command ensures that all the necessary configuration, including the package_spec, is correctly provided during the update process.

I hope this helps!

[Siddharthan-K]

Thanks for the clarification.

Since the Python SDK exposes agent_engines.update(), could you clarify its intended and supported use cases?

Specifically:

• What fields are safe to update using update()?
• What is package_spec and whether is it always required when calling it?