Deployment Fixes: Cloud Build and Agent model names

[AnkitGole007]

The previous build was failing before it even started. Installing pydantic-ai pulled in every optional add-on the package offers, none of which we use. One of those add-ons, the logging library, had a version conflict that made pip give up entirely and refuse to build the image.

This PR swaps the full package for a slimmer version that only includes the two things we need: Google model support and Vertex AI. The build now completes cleanly.

Agent model names

Two background agents: an example agent and a creative writing agent, were configured to use a model called gemini-3.1-flash-lite-preview. That model does not exist. If the message router ever sent a conversation to either of those agents, the request would fail with a 404. Both have been updated to use gemini-2.5-flash, the same model the WhatsApp agent uses.

Confirmed working end to end

After these fixes were deployed, a test message: "Hello B2, what is Givelight?" was sent through Postman directly to the /message endpoint. The service returned a 200 with a real response from the agent in 16 seconds.

POST /message → 200 OK · 16.12 s
{ "response": "Hello! Givelight is our programme designed to provide support
  to orphans and vulnerable children. We focus on ensuring they have access
  to education, healthcare, and a nurturing environment. How can I help you
  with Givelight today?" }

What was actually causing the 404

An earlier theory pointed to an IAM permission as the blocker. Eman dug into it and confirmed that was not the case, a missing permission would return a 403, not a 404. The real issue was that gemini-2.0-flash has been retired on Vertex AI in us-east1. Google returns "Publisher model not found" for any request that names a retired model, regardless of who is asking. gemini-2.5-flash works with the same credentials and returns a 200.

The whatsapp agent was already updated to gemini-2.5-flash in a previous PR. This build confirms the fix is live and working. The service is ready. The remaining step is for the webhook in benevolent-bandwidth to be pointed at the Cloud Run URL so live WhatsApp messages start flowing through.

[Copilot]

Pull request overview

This PR fixes deployment failures and runtime 404s by (1) tightening the API image dependency set to avoid installing unused pydantic-ai extras that were causing pip resolution conflicts, and (2) updating agent YAML configurations to use an existing Vertex AI Gemini model name.

Changes:

• Replace pydantic-ai with pydantic-ai-slim[google,vertexai] in the API requirements to avoid pulling unnecessary optional dependencies.
• Update the configured Gemini model for whatsapp_agent, example_agent, and creative_agent to gemini-2.5-flash.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
requirements-api.txt	Switches to pydantic-ai-slim with only Google/Vertex AI extras to prevent dependency-resolution failures during build.
agents/whatsapp_agent.yaml	Updates the WhatsApp agent model to gemini-2.5-flash.
agents/exampleAgent.yaml	Updates the example agent model to gemini-2.5-flash.
agents/creativeAgent.yaml	Updates the creative writing agent model to gemini-2.5-flash.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.