Manifesto: The Open Agentic Knowledge (OAK) repository
Join our community! Connect with contributors and users on Discord to discuss ideas, ask questions, and collaborate on the OAK repository.
Agents depend on APIs as much as AI. Chatbots chat, but agents act, and they act through APIs: checking calendars, booking flights, analyzing data, reconciling accounts, and controlling smart buildings. To realize this potential, we must build a common knowledge foundation that empowers AI to interact with the world's APIs reliably and without artificial barriers or unnecessary intermediaries.
Principles
-
Documentation Is Essential Infrastructure
While humans can work around incomplete API documentation through online research and experimentation, agent reliability depends on comprehensive documentation. API documentation is the map by which agents can navigate their action space. Good documentation was once a "nice-to-have", but is now a "need-to-have." -
Traditional API Integration Involved Middlemen
Before generative AI, the complexity of API integration gave rise to an ecosystem of intermediaries such as API marketplaces, aggregators, and integration platforms. These companies helped human developers integrate public APIs, while imposing additional costs and often creating vendor lock-in. -
AI Eliminates The Need For Intermediaries
Unlike human developers, AI doesn't require APIs to be simplified or aggregated — it just needs good documentation. This fundamental shift eliminates the need for middlemen, since AI can go direct. -
AI Can Generate Documentation
The same technology that demands better documentation also enables its creation. AI enables automatic pre-generation of high-quality documentation from code, web resources, PDFs, production telemetry, and even through experimentation. This will create a virtuous cycle of improvement. -
An Open Knowledge Layer Is Necessary
This virtuous cycle will be maximised by a collective open-source movement to establish a comprehensive "knowledge layer" for agents: a communal repository of detailed, AI-optimized information about every public API in the world. -
Standards Prevent Fragmentation
To avoid fragmentation and proprietary lock-in, we must build upon established, open standards. The OpenAPI Specification, owned by the Linux Foundation, already provides a robust foundation with its vibrant ecosystem of tools and services. -
Building on OpenAPI
The OpenAPI Initiative's Arazzo specification defines composable complex workflows built from OpenAPI operations, providing a timely declarative format needed for representing complex agentic tool knowledge. We will support these standards, and propose additional extensions as necessary to make them even more valuable to agents. -
MCP and OpenAPI Are Complementary
The future of AI needs both Model Context Protocol (MCP) and OpenAPI standards working in tandem. While OpenAPI and Arazzo provide the standard declarative format for representing agentic tool knowledge, MCP offers an ergonomic format for agents to access this knowledge, creating a complete ecosystem for AI interaction with APIs. -
Open Standards Ensure Interoperability
By embracing open standards like OpenAPI and Arazzo, we ensure AI agents can reliably discover, understand, and interact with APIs regardless of which model or platform they use, guaranteeing portability and longevity independent of any single company. -
Community Collaboration Accelerates Progress
This knowledge layer must be community-driven, welcoming contributions and feedback from both humans and AI to rapidly expand, refine and converge API knowledge. Together, we will build the most complete and AI-optimal representation of the world's APIs.
We invite developers, organizations, and AI builders to join this mission of creating an open-source knowledge layer for all the world's APIs, built on established open standards. Together, we will provide all agents with the knowledge necessary to reliably use the world's APIs, without vendor lock-in, risk of abandonment, or tolls paid to gatekeepers.
Note: This manifesto was created without the use of AI.