Awesome Agents on Google Cloud

A curated list of tools, docs, and resources for building AI agents on Google Cloud.

Start with the companion guide: The anatomy of an AI agent on Google Cloud: a complete guide, a map of the whole landscape with diagrams and a decision guide.

Frameworks

Agent Development Kit (ADK) - Google’s open-source, code-first agent framework; Python, Go, Java, TypeScript, and Kotlin.
adk-python - The Python SDK; the most mature of the five.
ADK 2.0 - The current generation: graph-based execution engine, parallel workflows, retries, human-in-the-loop pauses.
Agent Runtime framework support - LangGraph, LangChain, LlamaIndex, and AG2 deploy to the managed runtime too.

Agent Runtime - Managed agent runtime (formerly Vertex AI Agent Engine): sessions, memory, code sandbox, tracing, evals.
Cloud Run - Serverless containers; services, worker pools, jobs, and GPUs.
Cloud Run GPUs - NVIDIA L4 and RTX PRO 6000 Blackwell with scale-to-zero billing.
Agentic AI on GKE - The Kubernetes path for agent fleets and self-hosted models.
Agent Sandbox on GKE - Kernel-level isolation (gVisor) for agent-executed code.

Gemini API models - The current Gemini lineup, Pro to Flash-Lite.
Model Garden - 200+ models inside your Google Cloud project, including Claude, GLM, Kimi, and DeepSeek.
Gemma 4 - Google’s open model family, with function calling and structured output.
Gemma 4 function calling guide - How tool use works on Gemma 4.
Run Gemma on Cloud Run - Official pattern: Gemma 4 served with vLLM, wired into an ADK agent.
Agent Arena - Agent-specific leaderboard: tool orchestration and task completion across a million-plus real sessions.

Model Context Protocol (MCP) - The open standard for connecting agents to tools and data.
Google-managed MCP servers - Fully managed MCP endpoints for Google Cloud databases and services.
Google MCP catalog - Index of Google’s managed and open-source MCP servers.
MCP Toolbox - Open-source MCP server covering 40+ data sources.
Host MCP servers on Cloud Run - Run your own remote MCP server, serverless.
Agent2Agent protocol (A2A) - Agent-to-agent coordination; Linux Foundation governed, v1.0.
a2a-samples - Official multi-language A2A examples.

Agent Search - Turnkey retrieval over your data (formerly Vertex AI Search).
RAG Engine - Managed RAG pipeline with pluggable chunking, embeddings, and vector stores.
Vector Search - Raw vector infrastructure for billions-scale retrieval.
AlloyDB AI - Vector search and AI queries inside your operational database.
BigQuery vector search - Semantic search where your analytical data already lives.
Grounding with Google Search - Fresh public-web knowledge without retrieval infrastructure.

Sessions - Managed conversation state, callable from agents hosted anywhere.
Memory Bank - Long-term memory that extracts and consolidates facts instead of storing raw transcripts.
ADK sessions - Session-state backends: in-memory, your own Postgres, or managed Sessions.
ADK memory - Pluggable memory interface; Memory Bank is one backend among several.

Agent observability - OpenTelemetry GenAI traces rendered in Cloud Trace.
adk eval - Evaluate tool-call trajectories, not just final answers; runs in CI.
Agent Evaluation - Synthetic users and continuous scoring on live traffic.
Model Armor - Screens prompts and responses: prompt injection, jailbreaks, data loss.
Agent Identity - IAM principal type built for agents; non-impersonable, no long-lived keys.
ADK tool confirmation - Require human approval before a tool runs.
Secure AI Framework (SAIF) - Google’s risk map for agent security.

agents-cli - Scaffold, evaluate, and deploy agents with Terraform and staging-then-prod CI/CD.
adk-samples - 70+ sample agents across Python, TypeScript, Go, Java, and Kotlin.
agent-starter-pack - The predecessor to agents-cli; now in maintenance mode.
Agent Garden - In-console gallery of agent samples with one-click deploy.

Your first agent with ADK - From prototype to agent.
ADK agent with an MCP server on Cloud Run - Tools as their own services.
Multi-agent systems with ADK and A2A - Build and deploy a multi-agent system.
Kaggle 5-day AI Agents Intensive - Self-paced course with five whitepapers, from intro to production.
Build intelligent agents with ADK - Google Skills course.

The Agent Factory - Google Cloud DevRel’s agent podcast; episodes map closely to the layers above.

Contributions welcome. Keep entries current (post-2025 product names), one line each, with a short description of why the resource matters.