✨

Social Spark — Docs

The finished, governed reference implementation. Served straight from this repo — no build step, just point a static server at this folder.

Start here

Pre-camp · do this first
Attendee setup guide
GCP project, spend cap, local tools, Antigravity with the Agent Platform fallback, buddy check — everything to finish before the first Saturday.
4 phases · 45–60 min · screenshots included
Jump straight in
Lab 1 — Build →
The first workshop. Write a real ADK agent, end to end.
11 steps · ~2 hours

GDG DevCamp — the four pillars

The platform
The Gemini Enterprise Agent Platform: the Build, Scale, Govern and Optimise pillars and the services within each.
Diagram from Google Cloud — read the official overview ↗
How the labs work
One pillar at a time
Each lab takes a single pillar and applies it to this agent: Theory (the general concept), a hands-on Workshop, and a self-directed Assignment you do on your own agent afterwards.
Every lab names one deliberate error up front — a real constraint you will hit — rather than quietly routing around it.
Four sessions · Sept–Oct
Lab 1 · available
Build
ADK, Model Garden, Agent Skills, tools, MCP toolsets — design and compose the agent.
Session 1 · catch up with pillar-build
Lab 2 · available
Scale
Agent Runtime, Agent Identity, Sessions, Memory Bank — run the agent on managed infrastructure, and keep what matters when it scales to zero.
Session 2 · catch up with pillar-scale
Lab 3 · available
Govern
Model Armor, Cloud DLP, Governance Policies, Agent Registry, Agent Gateway — control what the agent may do.
Session 3 · catch up with pillar-govern
Lab 4 · available
Optimise
Golden evals, simulated users, BigQuery Agent Analytics, traces, eval-gated deploys — measure it, then make it better and cheaper.
Session 4 · catch up with pillar-govern

Reference — how this repo's production setup works

Agent Runtime deploy
One command, the /api passthrough, sessions & memory, frontend on Cloud Run
Cloud Run deploy (optional)
Optional reference: IAM, Dockerfile, eval-gated CI/CD
Observability
Cloud Trace, BigQuery analytics, what the table shows
Model Armor setup
Guardrails: templates, grants and switches
Injection attack table
What's blocked vs. not
Lab 3 demo script
Run of show, commands and fallbacks for the Session 3 lead
Lab 4 demo script
Run of show, commands and fallbacks for the Session 4 lead
Agent Engine RAG
Memory agent deploy
Worth reading before you deploy anything
Learnings
Every gotcha hit while building this, dated and explained — what happened, why, and what it means for your own agent. Most of them cost hours.