Google Agent Development Kit (ADK)

v2.0

Google

Agentmulti-agentopen-sourcegoogleframework
83
Strong
About This Agent

Open-source, code-first framework for building, evaluating, and deploying AI agents. Supports workflow agents, multi-agent hierarchies, built-in evaluation, and deployment to Vertex AI Agent Engine. Underlies Google's broader agent stack and is Gemini-optimized but model-agnostic.

Last Evaluated: June 10, 2026
Official Website

Trust Vector Analysis

Dimension Breakdown

🚀Performance & Reliability
+
task completion accuracy

Review of built-in evaluation tooling and reported agent benchmark workflows

Evidence
ADK Documentation - Evaluate — Built-in evaluation framework for assessing response quality and step-by-step trajectory against test cases
mediumVerified: 2026-06-10
tool use reliability

Tool integration testing across function, OpenAPI, and MCP tool types

Evidence
ADK Documentation - Tools — Rich tool ecosystem: function tools, OpenAPI tools, MCP tools, Google Cloud tools, and third-party library integrations
highVerified: 2026-06-10
multi step planning

Assessment of workflow agent primitives and graph-based orchestration in ADK 2.0

Evidence
ADK 2.0 Release Notes — ADK 2.0 GA adds graph-based, dynamic, and collaborative workflows alongside sequential, parallel, and loop workflow agents
highVerified: 2026-06-10
memory persistence

Memory and session service architecture evaluation

Evidence
ADK Documentation - Sessions and Memory — Session, state, and memory services with in-memory, database, and Vertex AI managed backends
mediumVerified: 2026-06-10
error recovery

Error handling pattern review and community issue analysis

Evidence
ADK GitHub Issues — Callbacks and loop agents enable retry patterns; robust recovery logic remains developer responsibility
mediumVerified: 2026-06-10
agent collaboration

Multi-agent hierarchy and delegation capability testing

Evidence
ADK Documentation - Multi-Agent Systems — First-class multi-agent hierarchies with delegation, agent-as-tool composition, and A2A protocol support
highVerified: 2026-06-10
🛡️Security
+
tool sandboxing

Security architecture review of tool execution paths

Evidence
ADK Documentation - Code Execution — Sandboxed code execution available via Vertex AI code executor; general tool sandboxing is developer responsibility
mediumVerified: 2026-06-10
access control

Access control and authentication capability assessment

Evidence
ADK Documentation - Authentication — Tool authentication framework (OAuth2, API keys, service accounts) plus Google Cloud IAM when deployed on Vertex AI
mediumVerified: 2026-06-10
prompt injection defense

Review of documented safety patterns and guardrail mechanisms

Evidence
ADK Documentation - Safety and Security — Documented guidance on guardrails, callbacks for input/output screening, and Gemini safety settings; no automatic injection blocking
mediumVerified: 2026-06-10
data isolation

Data architecture and session isolation review

Evidence
ADK Documentation - Sessions — Per-session state isolation with user- and app-scoped state prefixes; self-hosted deployments control data boundaries
mediumVerified: 2026-06-10
open source transparency

Source code and license review

Evidence
google/adk-python GitHub — Apache 2.0 license, fully open source with public development across Python, Java, TypeScript, Go, and Kotlin SDKs
highVerified: 2026-06-10
🔒Privacy & Compliance
+
data retention

Privacy architecture review across deployment options

Evidence
ADK Deployment Options — Self-hosted deployments give full control over retention; Vertex AI Agent Engine follows Google Cloud data governance
mediumVerified: 2026-06-10
gdpr compliance

Compliance capabilities assessment for framework and managed deployments

Evidence
Google Cloud Compliance — Vertex AI deployments inherit Google Cloud GDPR commitments; self-hosted compliance depends on operator configuration
mediumVerified: 2026-06-10
third party data sharing

Data flow analysis of model and tool integrations

Evidence
ADK Documentation - Models — Prompts and tool data are sent to the configured model provider (Gemini API by default, others via LiteLLM)
mediumVerified: 2026-06-10
local deployment option

Deployment options assessment including local model support

Evidence
ADK Documentation - Models — Model-agnostic via LiteLLM including Ollama and other local model backends; framework itself runs anywhere
highVerified: 2026-06-10
👁️Trust & Transparency
+
documentation quality

Documentation completeness review

Evidence
ADK Documentation — Comprehensive docs covering agents, tools, workflows, evaluation, deployment, and safety with quickstarts and samples
highVerified: 2026-06-10
execution traceability

Tracing and observability capabilities assessment

Evidence
ADK Dev UI and Evaluation — Built-in dev UI for step-by-step inspection of events, state, and tool calls; trajectory evaluation against expected steps
highVerified: 2026-06-10
decision explainability

Explainability features assessment

Evidence
ADK Events Model — Event stream exposes agent reasoning steps, function calls, and state deltas for inspection
mediumVerified: 2026-06-10
open source code

Open source assessment

Evidence
google/adk-python GitHub — Apache 2.0 licensed; announced at Cloud Next April 2025, Python v1.0 May 2025, ADK 2.0 GA May 2026
highVerified: 2026-06-10
community activity

Community engagement and release cadence analysis

Evidence
ADK GitHub Activity — Active development with frequent releases, multi-language SDK expansion, and growing contributor base since April 2025 launch
highVerified: 2026-06-10
⚙️Operational Excellence
+
ease of integration

Integration complexity assessment

Evidence
ADK Quickstart — pip-installable with code-first Pythonic API; agents definable in under 100 lines with CLI and dev UI included
highVerified: 2026-06-10
scalability

Deployment and scaling options assessment

Evidence
ADK Deploy to Agent Engine — Managed scaling via Vertex AI Agent Engine plus Cloud Run and GKE deployment paths
highVerified: 2026-06-10
cost predictability

Pricing model analysis

Evidence
ADK GitHub — Framework is free (Apache 2.0); costs come from model API usage and optional Vertex AI / Gemini Enterprise Agent Platform services
highVerified: 2026-06-10
monitoring capabilities

Monitoring features assessment

Evidence
ADK Observability Docs — Logging, tracing integrations, and evaluation tooling; Cloud Trace and third-party observability supported on managed deployments
mediumVerified: 2026-06-10
production readiness

Production readiness and release maturity assessment

Evidence
ADK Release Notes — Python v1.0 stable since May 2025; ADK 2.0 GA on 2026-05-19 with production-focused workflow and deployment improvements
highVerified: 2026-06-10
Strengths
  • +First-class multi-agent hierarchies, delegation, and A2A protocol support
  • +ADK 2.0 adds graph-based, dynamic, and collaborative workflows
  • +Built-in evaluation framework for response quality and trajectory testing
  • +Open source (Apache 2.0) with Python, Java, TypeScript, Go, and Kotlin SDKs
  • +Clean deployment path to Vertex AI Agent Engine, Cloud Run, or GKE
  • +Model-agnostic via LiteLLM while optimized for Gemini
Limitations
  • !Best experience is tied to the Google Cloud and Gemini ecosystem
  • !Tool sandboxing and injection defenses are largely developer responsibility
  • !Rapid release cadence has introduced breaking changes between major versions
  • !Managed features (Agent Engine, Gemini Enterprise Agent Platform) add paid cloud dependency
  • !Younger ecosystem than incumbent frameworks like LangChain/LangGraph
Metadata
license: Apache 2.0
supported models
0: Google Gemini (optimized)
1: Anthropic Claude
2: OpenAI models via LiteLLM
3: Local LLMs via LiteLLM/Ollama
programming languages
0: Python
1: Java
2: TypeScript
3: Go
4: Kotlin
deployment type: Self-hosted or managed (Vertex AI Agent Engine, Cloud Run, GKE)
tool support
0: Function tools
1: OpenAPI tools
2: MCP tools
3: Google Cloud tools
4: Third-party library tools
first release: 2025 (announced Cloud Next 2025-04-09; Python v1.0 May 2025; ADK 2.0 GA 2026-05-19)
pricing: Free framework (Apache 2.0); paid when using Vertex AI Agent Engine or Gemini Enterprise Agent Platform plus model API costs

Use Case Ratings

customer support

Strong fit for multi-agent support systems; underlies Google's Customer Engagement Suite agent stack

code generation

Capable via code execution tools and workflow agents, though not a purpose-built coding agent

research assistant

Multi-agent hierarchies with Google Search grounding work well for research pipelines

data analysis

Good with code executors, BigQuery and Google Cloud tool integrations

content creation

Workflow agents support multi-stage drafting and review pipelines

financial analysis

Viable on Vertex AI with enterprise controls; compliance hardening is the builder's responsibility

healthcare

Requires significant compliance work; Google Cloud HIPAA-eligible services help when deployed on Vertex AI