Looker ai agent

[ferraricharles]

The idea here is to have a way to implement the basic agent in dialogflow capable to offer an interface using the new capabilities of the Conversational Analytics API for Looker.

I've improved the way we instantiate the cloud build given that it now uses the Compute Engine service account in some cases, which would require us to grant broad IAM permissions to the compute SA, which wouldn't be safe at all, so I decided to create a new service account instead and use that for the cloudbuilds.

I've still used the compute engine service account for cloud run, giving it access to conversational analytics API and some other required AI features.

The cloudbuild instantiates a new cloudbuild using a DockerFile which will then create the container to be used by cloud run, proceeding finally with the infrastructure creation. This step was necessary to ensure we can get latest container.

The Python code is then responsible to communicate with the Looker Conversational Analytics API (currently in private preview), extract the textual response and assemble a link to the corresponding explore in Looker (to allow users to keep exploring).

The pieces required to be performed by the user are:
1 - Setup a Secret named with the following attributes:
{
"LOOKER_CLIENT_ID": "YOUR_LOOKER_CLIENT_ID",
"LOOKER_CLIENT_SECRET": "YOUR_LOOKER_CLIENT_SECRET",
"LOOKER_INSTANCE": "YOUR_LOOKER_INSTANCE_URL",
"LOOKML_MODEL": "your_lookml_model_name",
"LOOKML_EXPLORE": "your_explore_name"
}

2 - Enter the created agent and build their own Playbook as well as indicate to the agent that conversations should be started by that playbook. Currently playbooks can't be terraformed but this allows users to craft their own instructions to the agent and chose the LLM to be used (and potentially create and connect multiple tools).

Architecture:

Working Agent:

Connected to Slack:

Using Conversational Manager:

[ferraricharles]

Putting this back in draft to perform this migration in the code: https://cloud.google.com/gemini/docs/conversational-analytics-api/migration-guide

[fellipeamedeiros]

LGTM