Google Agent Development Kit
adk.dev
What it is
Google Agent Development Kit (ADK) is an open-source framework for building, debugging, evaluating, and deploying AI agents. It is available in Python, TypeScript, Go, Java, and Kotlin. Graph-based workflows can combine deterministic code with adaptive AI reasoning, and hierarchical agents can delegate work to specialized agents. ADK supports Gemini and adapters for other model providers, including locally running models. Its integrations catalog includes tools and connectors for MCP, Google Cloud, databases, search, observability, and third-party services. Teams can create multi-turn evaluation datasets and run evaluations locally through the CLI or development UI. ADK also supports live agents with bidirectional text and audio streaming, manages session context, and connects to longer-term memory services. Deployment options include Google Cloud Agent Runtime, Cloud Run, GKE, container-friendly infrastructure, and local Docker or Podman environments. Security guidance covers identity, authorization, guardrails, sandboxed code execution, and network controls. ADK is free; model and cloud-service use may require separate access or billing. Agent Config is narrower: it currently supports only Gemini models and requires Python.
Spec sheet
- What it does
- ADK is Google's open-source framework for building, debugging, evaluating and deploying production AI agents and multi-agent systems.Read 15 Oct 2026 · adk.dev
- Purpose
- ADK is an open-source framework for building, debugging, evaluating, and deploying AI agents.Read 4 Oct 2026 · adk.dev
- Deployment
- Agents can run locally, in containers, on Cloud Run, on GKE or on Google's managed Agent Runtime.Read 15 Oct 2026 · adk.dev
- Integrations
- The integrations catalog includes tools and connectors for MCP, Google Cloud, databases, search, observability, and third-party services.Read 4 Oct 2026 · adk.dev
- Security
- The security guidance covers identity and authorization, input and output guardrails, sandboxed code execution, and network controls.Read 4 Oct 2026 · adk.dev
- Support
- The site links to community resources and says Google and the ADK team do not support the content in external community resources.Read 4 Oct 2026 · adk.dev
- License
- The Python ADK repository is licensed under the permissive Apache License 2.0.Read 15 Oct 2026 · github.com
- Agent orchestration
- The framework supports modular multi-agent composition, delegation and routing between specialized agents.Read 15 Oct 2026 · developers.googleblog.com
- API surface
- Official references include Python, TypeScript, Go, Java and Kotlin APIs plus CLI, configuration and REST API references.Read 15 Oct 2026 · adk.dev
- CLI
- Agents CLI can scaffold, build, test, evaluate and deploy ADK agents from an AI-enabled development environment.Read 15 Oct 2026 · adk.dev
- Company founded
- Google was founded in 1998 by Larry Page and Sergey Brin.Read 15 Oct 2026 · en.wikipedia.org
- Company legal identity
- Google LLC is Alphabet's technology subsidiary and is headquartered in Mountain View, California.Read 15 Oct 2026 · en.wikipedia.org
- Company size
- LinkedIn lists Google as a public software-development company with 10,001-plus employees and headquarters in Mountain View, California.Read 15 Oct 2026 · linkedin.com
- Context and memory
- ADK manages session context and provides integration points for longer-term memory services.Read 4 Oct 2026 · adk.dev
- Context management
- ADK manages sessions, memory, events, tool outputs and artifacts, including summarization and context compression controls.Read 15 Oct 2026 · adk.dev
- Developer UI
- The local ADK web interface lets developers chat with agents and inspect events, state and execution traces step by step.Read 15 Oct 2026 · adk.dev
- Documentation depth
- The documentation site includes getting-started guides, component references, deployment, observability, evaluation, safety, MCP, A2A and API references.Read 15 Oct 2026 · adk.dev
- Evaluation
- Documentation covers evaluation criteria, user and environment simulation, custom metrics and optimization workflows.Read 15 Oct 2026 · adk.dev
- Graph workflows
- ADK 2.0 adds graph-based workflows that combine deterministic code with model reasoning and explicit execution paths.Read 15 Oct 2026 · adk.dev
- Installation
- The installation guide provides package installation instructions for all five supported languages.Read 4 Oct 2026 · adk.dev
- Installation requirement
- The Python quickstart requires Python 3.10 or later and installs the package with pip install google-adk.Read 15 Oct 2026 · github.com
- Integration ecosystem
- Google lists integrations spanning Google Cloud services, databases, observability platforms, communication tools, productivity apps and many third-party agent tools.Read 15 Oct 2026 · adk.dev
- Languages
- Official implementations are provided for Python, TypeScript, Go, Java and Kotlin.Read 15 Oct 2026 · adk.dev
- Launch news
- Google announced ADK on April 9, 2025 at Google Cloud NEXT as an open-source framework for production agent and multi-agent development.Read 15 Oct 2026 · developers.googleblog.com
- Limit: beta components
- ChatForest reports that Python 2.0 and Kotlin 0.1.0 were beta releases and that ADK for Android was not production-ready at review time.Read 15 Oct 2026 · chatforest.com
- Limit: ecosystem maturity
- ChatForest says ADK's community, tutorials and production case studies still lag behind LangGraph.Read 15 Oct 2026 · chatforest.com
- Limit: Google Cloud dependence
- Independent reviewers say the managed deployment path is Google Cloud-specific, creating friction for cloud-agnostic teams.Read 15 Oct 2026 · chatforest.com
- Limit: learning curve
- InfoWorld reports that ADK's extensive capabilities require substantial time and effort to learn.Read 15 Oct 2026 · infoworld.com
- MCP and A2A
- ADK supports MCP for agent-to-tool connectivity and A2A for agent-to-agent communication across frameworks.Read 15 Oct 2026 · javisreview.com
- Model support
- ADK is optimized for Gemini but provides adapters for other hosted and locally running model providers.Read 15 Oct 2026 · adk.dev
- Models
- ADK supports Gemini and adapters for other model providers, including locally running models.Read 4 Oct 2026 · adk.dev
- Multi-agent
- Agents can be arranged hierarchically and delegate tasks to specialized agents.Read 4 Oct 2026 · adk.dev
- Notable limitation
- Agent Config currently supports only Gemini models and requires Python.Read 4 Oct 2026 · adk.dev
- Praise: architecture
- Independent reviewers praise ADK's multi-agent architecture, graph workflows, language coverage and built-in developer tooling.Read 15 Oct 2026 · infoworld.com
- Praise: documentation and UI
- An independent course review highlights the local development UI and says evaluation and observability receive unusually substantial documentation.Read 15 Oct 2026 · aicoursereview.com
- Pricing reality
- Javis reports that the framework is free, but model usage, self-managed infrastructure and managed Agent Runtime consumption create operating costs.Read 15 Oct 2026 · javisreview.com
- Rating on Awesome Agents
- Awesome Agents rated ADK 7.6 out of 10 from its editorial review.Read 15 Oct 2026 · awesomeagents.ai
- Rating on ChatForest
- ChatForest rated ADK 3 out of 5 from its review.Read 15 Oct 2026 · chatforest.com
- Rating on InfoWorld
- InfoWorld's reviewer called ADK capable and mostly complete, while noting a significant learning curve and fewer examples for TypeScript, Go and Java.Read 15 Oct 2026 · infoworld.com
- Security guidance
- Official documentation has a dedicated safety and security section covering security considerations for AI agents.Read 15 Oct 2026 · adk.dev
- Self-hosting
- The framework can be containerized and operated on infrastructure chosen by the developer without changing agent code.Read 15 Oct 2026 · adk.dev
- Streaming
- The framework includes bidirectional audio and video streaming for multimodal conversations.Read 15 Oct 2026 · developers.googleblog.com
- Tools
- Agents can use built-in search and code-execution tools, MCP tools, OpenAPI tools, custom functions and third-party libraries.Read 15 Oct 2026 · developers.googleblog.com
- Voice
- ADK supports live agents with bidirectional text and audio streaming.Read 4 Oct 2026 · adk.dev
- Workflow patterns
- Built-in workflow agents support sequential, parallel and loop execution patterns.Read 15 Oct 2026 · developers.googleblog.com
- Workflows
- ADK supports graph-based workflows that combine deterministic code with adaptive AI reasoning.Read 4 Oct 2026 · adk.dev
Plans and prices
Strong points and limits
Strong points
- Available in five programming languages
- Supports hierarchical agents and delegated tasks
- Includes local evaluation tools
- Supports live text and audio streaming
- Offers cloud, container, and local deployment options
Limits
- Agent Config supports only Gemini models
- Agent Config requires Python
- Model and cloud usage may require separate billing
Who it suits
ADK suits developers building agents across supported languages and deployment environments, including multi-agent and voice workflows. Teams considering Agent Config should note its Gemini-only and Python requirements.
Verdict
ADK covers agent development from workflows and evaluation through deployment, with several language and infrastructure options. The framework is free, but model or cloud service use may carry separate access or billing requirements.
Where Specifiction places it
- AI Agent Platforms#7 of 738.8