google/adk-go

What it solves

Building reliable, maintainable AI agents can be messy and unstructured. ADK for Go brings software-engineering discipline to agent creation: you define agents, their tools, and how they orchestrate in regular Go code, so you get versioning, testability, and flexibility instead of ad‑hoc glue scripts.

How it works

You write agents as Go code using the ADK API. Each agent can call a rich ecosystem of pre‑built tools, custom functions, or existing integrations, and you compose multiple specialized agents into larger multi‑agent systems. The framework is model‑agnostic (though tuned for Gemini) and deployment‑agnostic, so the same agent code can be containerized and run on cloud platforms like Google Cloud Run. Because it's code‑first, all logic lives in your version control and can be tested like any Go module.

Who it's for

Go developers who want to build ambitious, cloud‑native AI agent applications and already value Go's concurrency, performance, and simplicity. It suits teams that prefer code over configuration and want to apply familiar software engineering habits to agent development.

Highlights

• Idiomatic Go that feels natural and leverages Go's strengths.
• Rich tool ecosystem: pre‑built tools, custom functions, or existing integrations.
• Code-first design for flexibility, testability, and versioning.
• Modular multi‑agent systems for scalable applications.
• Deploy anywhere, with strong cloud‑native support (e.g., Cloud Run).
• Model‑agnostic and compatible with other frameworks.