Developers everywhere face a common challenge: it’s getting easier and easier to prototype an AI agent, but harder to turn that prototype into something reliable, testable, and ready for production. Orchestrating LLM interactions, managing state, wiring up function calling, integrating multiple tools, evaluating performance, tracing failures, and deploying to production often require complex custom code and scattered tooling.
The DigitalOcean Gradient™ AI Agent Development Kit (ADK), now available in public preview, is designed to solve exactly that problem. It’s a code-first SDK for building, testing, and deploying multi-step agent workflows from your existing development environment, and it works with popular Python-based agent frameworks like LangGraph, LangChain, CrewAI, as well as custom-built agent systems. The ADK provides built-in support for evaluations, traceability, and knowledge bases so you can move from prototype to production with confidence.
To see the ADK in action and for help getting started, check out our tutorial.
Why we built the ADK
Most AI frameworks solve the “day-one” problem: how to create a working agent. But they fall short on the “day-two” problem: how to operate that agent in production.
We built the ADK to give you a standardized, production-grade framework that handles the full lifecycle:
- Orchestration: Build multi-step workflows without boilerplate
- State management: Keep track of what your agent is doing across steps
- Tool integration: Register custom functions/APIs as first-class tools
- Knowledge base support: Connect agents to your existing DO Knowledge Bases
- Evaluations: Measure correctness, security, tone, and retrieval quality
- Tracing: Understand how your agent behaves at every span and step
- Deployment: Ship your entire agent system, from logic to KBs to tools, with one command
The result is a code-first experience that lets you move from local development → observability → evaluation → deployment in a single, consistent workflow.
What’s included in the public preview
Our goal is to provide a strong foundation for developers to build, test, and deploy AI agent workflows directly from their development environment. In the private preview, you could leverage the core SDK framework with CRUD APIs, connect models, organize projects in workspaces, use the entry point decorator and CLI, and take advantage of MCP support on the Agent Platform.
Building on this foundation, the public preview introduces a set of powerful new features to help you run production-level workflows more efficiently and with deeper visibility:
- Traces & Insights for Any Workflow: Add full tracing to your agent logic using custom decorators across popular Python-based agent frameworks like LangGraph, LangChain, CrewAI, or even fully custom workflows. If you’re using LangGraph, traces are captured automatically with zero additional configuration.
- Knowledge Base (KB) Support: Attach existing DigitalOcean Knowledge Bases to your agents so workflows can reference domain-specific content. This ensures your agents have reliable context without extra setup.
- Evaluations for Multi-Step Agents: Build test cases from CSV datasets, apply DO-supported metrics, and evaluate entire multi-step workflows. Step-by-step results appear in the UI or CLI and link directly to execution traces, making debugging easier.
- A2A (Agent-to-Agent) Communication (Coming Soon): Expose agents as A2A-compliant endpoints, auto-generate AgentCards, and trace multi-agent interactions. Internal multi-agent calls are supported in this preview, while external calls will arrive in a future release. Core protocol, security, and logging are included.
- Easy Deployment: Define your agent in code and deploy it with a single command
gradient agent deploy. No more deploying each resource individually.
Try the Agent Development Kit today
The ADK makes it easier to build, test, and deploy AI agents without worrying about all the setup and boilerplate. You can start creating your first agent today, run it locally, connect it to your Knowledge Bases, trace and evaluate its behavior, and deploy it with a single command.
Not sure where to get started? Check out our ADK repository on Github for templates on building and deploying agets using DigitalOcean’s Gradient AI ADK. Each template demonstrates a different agent architecture and can be used as a starting point for ADK deployments.
To get started with the ADK, enable the public preview on your Feature Preview page in the DigitalOcean Cloud Console. Once you’ve opted in, access will be granted in approximately 10–15 minutes.