adk-python
Agent Development Kit (ADK) is Google's open-source, code-first framework for building, evaluating and deploying AI agents. It applies ordinary software engineering practices to agents: you define agents, tools and orchestration in Python, test them, version them and deploy them like any other service. ADK is optimised for Gemini but is model-agnostic and deployment-agnostic, and it is also available for Java, Kotlin, Go and TypeScript.
Version 2.0 adds a graph-based workflow runtime for deterministic flows with routing, fan-out and fan-in, loops, retries, state and human-in-the-loop steps, plus a Task API for structured delegation between agents. You can compose specialised agents into hierarchies, give them pre-built tools, custom functions, OpenAPI specs or MCP tools, guard sensitive tools with explicit confirmation, and even define agents without code through Agent Config. A built-in web UI (`adk web`) helps you test, debug and evaluate agents, and deployment targets include Cloud Run and Vertex AI Agent Engine.
Getting started: requires Python 3.11+. Install with `pip install google-adk` (Google recommends its constraints files for dependency safety), create an agent, then run `adk web` to try it in the browser.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-10-24
- Commits, last 90 days
- 100+
- Releases, last 90 days
- 18
- Latest release
- v2.13.0 (2026-10-23)
- Open issues
- 402
- Recent issues with at least one comment
- 15/20
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #6099 Decision ledger: persist the authority, refusal, and approval behind each tool call48 comments · 2026-06-13
- #3343 Update database schema automatically23 comments · 2025-10-30
- #7464 Redis session keys can collide for distinct identity tuples20 comments · 2026-10-08
- #7384 A task-mode root agent loses its task input once it asks the user a question14 comments · 2026-10-02
- #7351 Live inference_call_count_v1 changes with text chunk boundaries12 comments · 2026-09-30
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You want Google's Apache-2.0, code-first toolkit with graph workflows, multi-agent hierarchies, a built-in dev UI, and evaluation tooling.
- Your team deploys on Google Cloud (Cloud Run or Vertex AI Agent Engine) or works across Python and ADK's other language editions.
- Your infrastructure is centred on Azure and .NET; Microsoft Agent Framework is tailored to that stack.
- You want a visual low-code workflow builder rather than writing Python code; n8n or Activepieces fit better.
How it differs
vs agent-framework★ 14.3kGoogle ADK (Apache-2.0) leans towards Gemini and Vertex AI / Cloud Run deployments; Microsoft Agent Framework (MIT) provides first-class Python and .NET APIs oriented towards Azure and Microsoft Foundry.
vs Pydantic AI★ 20.9kGoogle ADK includes a built-in dev UI, evaluation suite, and multi-agent graph hierarchies; Pydantic AI focuses on lightweight type-safe tool definitions and Pydantic output validation.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!