Built on Google

How it fits together on Google Cloud

WHO USES IT GOOGLE AI AND SERVICES Marketing user chats in the browser about 10 people RB2B weekly file website visitors (CSV) uploaded in the chat Gemini Enterprise the company AI app later: the front door Cloud Scheduler later: weekly run, leads emailed out Google sign-in IAP: named company users only MARKETING'S GOOGLE CLOUD PROJECT · PRIVATE Cloud Run · one container The agent · built with Google ADK understands the request, picks the tool, asks Gemini to write a short answer The tool · lead check (fixed rules, no AI) 4 checks: Who · Which · What · Rule out points, top 3 people per account, account notes, Excel workbook Editable lists 13 lists Marketing controls: accounts, pages, competitors… Memory only the file and workbook, cleared after 30 idle minutes; never saved Own service account: it may only call Gemini and write logs. Built and changed only through reviewed code (Terraform). answer + workbook A2A Gemini on Vertex AI writes the summary and answers follow-ups gets counts and titles only Central AI gateway next: picks the best Gemini model per call Cloud Logging errors and audit only, no file content Later data sources Salesforce: already a contact? Pardot: content-download leads BigQuery: lists and results
Solid: live in the betaDashed: planned for the full versionPersonal data stays inside the project, in memory only
Vertex AI Agent Enginethe program's default agent runtime; being compared with Cloud Run before the move to the business unit's projects
Cloud Build + Artifact Registrypackage the code and lists into a container; never any data
Terraform on Google Cloudthe business unit's projects are built only from reviewed code