AI agents have become core to many modern applications – automating workflows or orchestrating complex tasks. However, a key challenge persists – fragmentation. Today’s developers rely on frameworks such as AutoGen, LangGraph, and CrewAI, each with its own strengths but lacking a shared foundation. This fragmentation can make it difficult to port agents between frameworks, reuse components, and scale solutions across platforms.
To address these challenges, Oracle is introducing Open Agent Specification (Agent Spec) – an open, standardized representation that brings portability, reusability, and extensibility to the AI agent ecosystem.
Agent Spec is a framework-agnostic declarative specification designed to make AI agents and workflows portable, reusable, and executable across any compatible framework. Inspired by the success of representations like Open Neural Network Exchange (ONNX) for ML models, Agent Spec aims to bring that same level of interoperability and optimization to the AI agent space to foster an ecosystem of tools to develop on top of it.
Key Benefits
With Open Agent Specification, developers can:
-
Define agents once and run them flexibly nearly anywhere – from local development environments to cloud-based runtimes.
-
Build standalone agents or agentic workflows that can interface with external tools, data sources, and LLMs in a consistent and declarative manner.
-
Choose the preferred execution runtime without rewriting agent logic.
-
Leverage a standardized, versioned representation to simplify maintenance and collaboration.
With Open Agent Specification, an Enterprise can:
-
Seamlessly scale AI agents from prototype to production without costly rewrites or reengineering.
-
Reduce lock-in to specific technology artifact or deployment platform.
-
Maintain AI agent deployment via simple configuration update rather than heavy software upgrade.
-
Exchange AI agents among different internal organizations or with external business partners.
Much like how ONNX helped unify model exchange across ML libraries, Agent Spec bridges agent development fragmentation, accelerating the path to open, composable, and intelligent agent systems.
Built to Evolve with the Agent Ecosystem
Agent Spec is built with and for the community – designed to evolve through real-world use, extensible modules, and support for custom components. Agent Spec provides strong synergies with various standardization efforts like Model Context Protocol (MCP). By supporting MCP servers, Agent Spec enables AI agents to access external APIs through a standardized tool interface (understand your organization’s data and privacy policies before using) – reducing the need for custom adapters for every external API and streamlining tool integrations.
To drive adoption and collaboration, we are open sourcing:
-
Open Agent Specification – A declarative schema for defining agents and workflows. A detailed technical report is available as well for interested readers.
-
Agent Spec SDK – Tools to build, validate, and transform Agent Spec definitions.
-
Agent Spec Reference runtime – WayFlow as a reference runtime for Open Agent Specification.
Execution frameworks (e.g., WayFlow, AutoGen, LangGraph) can implement Agent Spec runtime adapters to seamlessly interpret and orchestrate agents defined in Agent Spec.
Oracle Leads the Way in Adoption
As we open-source Open Agent Specification, several Oracle products are already adopting it to build interoperable, production-grade AI agents. For example, Oracle Applied AI, Oracle Financial Services Global Industries, and Oracle Autonomous Database Select AI are integrating Agent Spec configurations to enable enterprise agents. Agent Spec configurations are planned to be supported on other Oracle platforms in future, including Oracle Cloud Infrastructure.
Quote from Sandesh Rao, Vice President, Applied AI Technologies
“With Open Agent Specification and its runtime (WayFlow) integrated in Oracle Private Agent Factory, and in collaboration with Oracle Labs, we are making it easier than ever for enterprise customers to design and deploy production-grade AI agents.”
Quote from Jason Wynne, Senior Vice President, Financial Services Global Industries Unit
“With multiple AI Agent use-cases automating financial crime investigations already in production, in close collaboration with Oracle Labs and leveraging WayFlow as Open Agent Specification runtime, we see Agent Spec as a critical enabler for rapidly integrating our agents into different Oracle platforms. This shared foundation accelerates our ability to deploy, manage, and enhance Agentic solutions for financial services compliance and risk mitigation.”
Quote from Mark Hornick, Senior Director, Oracle Database Machine Learning and AI Product Management
“Enterprise users need a convenient and efficient way to exchange and deploy their agentic solutions. Given its focus on enabling database-centric AI-driven apps and reducing time-to-market, Select AI is also adopting Open Agent Specification – enabling interoperability and reusability of agents across platforms. In collaboration with Oracle Labs, we are helping to empower users to seamlessly integrate, manage, and easily enhance enterprise agentic applications.”
Contribute to a shared foundation for AI agents. Start exploring with Open Agent Specification today to create, share, and scale agents across frameworks. We also provide a full technical report for interested readers.
👉 Check out Open Agent Specification and WayFlow examples to get started.