Building AI Agents with the GO Agent Development Kit (ADK) — 2026 Edition (v2)

Google's Go Agent Development Kit (ADK) v1.5.1 has been updated with production hardening, dependency upgrades, and multi-agent delegation support via the A2A protocol. The toolkit enables developers to build, run, and deploy native AI agents in Go, with improvements including standardized logging, graceful shutdown handling, and cross-submodule CI/CD automation.

This updated edition incorporates several key architectural improvements, production hardening, and dependency upgrades over the initial release: v0.2.0 to v1.5.1 google.golang.org/adk , delivering enhanced model context handling and agent execution pipelines. v1.65.0 google.golang.org/genai with native Vertex AI authentication auto-detection. 1.26.3 go 1.25.0 module directives across all sub-modules. github.com/a2aproject/a2a-go v0.3.15 with sub-modules a2a-client-go , a2a-server-go , a2a-master-go , a2a-gemini3-go for multi-agent delegation. slog Standardization slog with JSONHandler across all entrypoints for native Google Cloud Logging compatibility. signal.NotifyContext context.Background , os.Interrupt across all sub-module entrypoints to handle container termination cleanly. rollDieTool sides <= 0 to prevent runtime panics from malformed LLM tool calls. CMD Generation cloudrun.go to construct container arguments with json.Marshal , ensuring valid CMD syntax regardless of feature flags. Makefile to format go fmt , lint go vet , test go test , build, and clean across all 5 submodules hello-agent , a2a-client-go , a2a-gemini3-go , a2a-master-go , a2a-server-go . go test -v ./... , formatting go fmt , linting go vet ./... , and make build pass with 0 errors.This tutorial provides a comprehensive guide to building, running, and deploying native AI Agents using the Go programming language and the official Go Agent Development Kit ADK google.golang.org/adk . Go Golang is an open-source programming language created at Google. Renowned for its simplicity, concurrency support via goroutines, and fast execution speed, Go is an exceptional choice for building high-throughput microservices and distributed agent systems. This project has been updated to run on Go 1.26 specifically go1.26.3 linux/amd64 with go 1.25.0 module directives across all sub-modules.If you are using gvm to manage Go installations: Dependencies for Linux / Debian / Ubuntu sudo apt-get update sudo apt-get install -y bison curl git mercurial make binutils gcc build-essential Install GVM if not already installed bash < < curl -s -S -L https://raw.githubusercontent.com/moovweb/gvm/master/binscripts/gvm-installer Install and switch to Go 1.26.3 gvm install go1.26.3 -B gvm use --default go1.26.3 Verify that your active Go binary is Go 1.26.3 : bash $ go version go version go1.26.3 linux/amd64 go 1.25.0 All go.mod files in this workspace have been upgraded from go 1.24.4 to go 1.25.0 to support modern language features and telemetry packages required by Go ADK v1.5.1 . The Go Agent Development Kit ADK google.golang.org/adk is a code-first toolkit developed by Google for building, evaluating, and deploying sophisticated AI agents. github.com/a2aproject/a2a-go v0.3.15+ for multi-agent collaboration.Official Repositories & Docs: go.mod The updated agent stack relies on the latest releases: module hello-agent go 1.25.0 require google.golang.org/adk v1.5.1 google.golang.org/genai v1.65.0 github.com/a2aproject/a2a-go v0.3.15 Clone the repository and set up environment credentials: git clone https://github.com/xbill9/adk-hello-world-go cd adk-hello-world-go Run init.sh to initialize settings: bash $ source init.sh --- Authentication Method --- Do you want to use a Gemini API Key for authentication? y/n : n --- Setup complete --- Then load project-wide environment variables using set env.sh : bash $ source set env.sh --- Setting Google Cloud Environment Variables --- Checking gcloud authentication status... gcloud is authenticated. Exported PROJECT ID=comglitn Exported PROJECT NUMBER=1056842563084 Exported GOOGLE CLOUD PROJECT=comglitn Exported GOOGLE GENAI USE VERTEXAI=TRUE Exported GOOGLE CLOUD LOCATION=us-central1 --- Environment setup complete --- If Google Cloud authentication expires, refresh credentials: gcloud auth login gcloud auth application-default login The project includes unified Makefile targets for linting, formatting, compiling, and testing all 5 agent submodules hello-agent , a2a-client-go , a2a-gemini3-go , a2a-master-go , a2a-server-go : Format code across all submodules make format Lint code with go vet across all submodules make lint Run unit tests across all submodules make test Build executables across all submodules make build Clean build artifacts make clean slog All entrypoints use slog configured with JSONHandler to emit logs compatible with Google Cloud Logging: logger := slog.New slog.NewJSONHandler os.Stdout, nil slog.SetDefault logger Service entrypoints use signal.NotifyContext to handle SIGINT and SIGTERM signals gracefully during Cloud Run container lifecycle events: ctx, cancel := signal.NotifyContext context.Background , os.Interrupt defer cancel Custom agent tools validate input bounds to prevent runtime panics when executing tool calls generated by LLMs: func rollDieTool tc tool.Context, args rollDieToolArgs int, error { if args.Sides <= 0 { return 0, fmt.Errorf "number of sides must be greater than 0, got %d", args.Sides } return rand.Intn args.Sides + 1, nil } Run the agent interactively in your terminal using cli.sh : php $ source cli.sh go run agent.go User - what can you do? Agent - I can help you with a variety of tasks I can answer questions, search information on Google, and tell you current time or weather in any city. User - what is the weather in NYC? Agent - The current weather in New York City is sunny with a temperature of 65°F 18°C ... Launch the embedded Web UI with web.sh : bash $ source web.sh Navigate to http://localhost:8081 in your browser to interact with the visual chat interface. The 2026 update introduces full Agent-to-Agent communication modules: hello-agent : Core single-agent implementation. a2a-server-go : Exposes agent capabilities over A2A HTTP endpoints. a2a-client-go : Client library for dispatching requests to remote A2A agents. a2a-master-go & a2a-gemini3-go : Orchestrators for complex multi-agent workflows using Gemini models. quickrun.sh Deploy to Google Cloud Run using the automated build script: bash $ source quickrun.sh Building using Dockerfile and deploying container to Cloud Run service adk-hello-world-go ... ✓ Building Container... ✓ Creating Revision... ✓ Routing traffic... Service URL: https://adk-hello-world-go-1056842563084.us-central1.run.app adkgo CLI Deployment Build the adkgo CLI tool from the adk-go upstream repository: cd ~ git clone https://github.com/google/adk-go cd adk-go go build ./cmd/adkgo Deploy using adkgo ./adkgo deploy cloudrun \ -p $GOOGLE CLOUD PROJECT \ -r $GOOGLE CLOUD LOCATION \ -s adk-hello-world-go \ -e ./hello-agent/agent.go proxy.sh Connect to your deployed Cloud Run service locally: bash $ source proxy.sh PROXY URL http://127.0.0.1:8081/ui/?app=hello time agent Proxying to Cloud Run service adk-hello-world-go in region us-central1 In this updated version: v1.5.1 v1.65.0