APIs make Agents Tick

AI agents are as much about APIs as they are about LLMs. Chatbots chat, but agents act. Through APIs they can check your calendar, book your flights, analyze your data, reconcile your accounts, or control your smart home. APIs transform AI from chatbots into genuine digital assistants that get things done.

Before GenAI, working with APIs was a chore. This spawned an ecosystem of intermediaries: API marketplaces, API aggregators, and integration platforms as a service (iPaaS). These intermediaries simplified API integration for human developers, but were an extra middleman, imposing additional direct costs and sometimes vendor lock-in.

But AI doesn't need middlemen. It just needs good documentation.

API documentation is need-to-have

Good API documentation used to be a nice-to-have. Developers appreciated it, but were paid to muddle through with incomplete docs, community forums, and trial and error.

For agents, documentation is a need-to-have. It's essential infrastructure. Documentation is the map by which the agent navigates the API landscape.

Luckily, AI itself offers the solution to the documentation problem. We can now use AI to automatically generate documentation at scale from code, web resources, PDFs, production telemetry, and even agentic experimentation. The technology that demands better documentation also enables us to create it.

The agentic knowledge layer

It is now time for an open-source movement to establish a "knowledge layer" for agents: a communal repository of detailed, AI-optimized knowledge about every public API in the world.

We must move quickly to avoid a proprietary alternative gaining dominance, whether through closed-source technology or a proprietary standard. And we must move cohesively to avoid fragmentation, by working with established standards. An effective and widely adopted standard for representing API knowledge already exists: OpenAPI. This is governed by the Linux Foundation and has a vibrant ecosystem of tools and services.

Recently, the OpenAPI Initiative launched v1 of Arazzo, which defines workflows that can be composed from OpenAPI operations and other Arazzo workflows. We are aligning with Arazzo as the standard declarative format for agentic tool knowledge.

We are also proposing extensions to the OpenAPI Initiative's standards to help represent specific information to help AI agents make better decisions about which API operation or workflow to use and how to reliably call it, and interpret the response.

MCP and OpenAPI are Complementary

The future of AI needs both Model Context Protocol (MCP) and OpenAPI/Arazzo. Where OpenAPI and Arazzo focus on providing the standard declarative format for representing agentic tool knowledge, MCP provides an ergonomic format for agents to access this knowledge.

This open-source project is focusing solely on the knowledge layer, which may be consumed by agents in multiple ways, including via MCP. We're embracing the OpenAPI Initiative's standards as the correct and established standard for representing this knowledge.

Jentic has a commercial MCP service (with a free tier) that allows easy discovery and invocation of API operations and workflows represented in this open-source knowledge layer.

Community focus

We are investing in the OAK repository with the aim of creating open-source agentic knowledge for all the world's APIs. We welcome everyone to join us in this mission of accelerating agent capabilities and reliability, while cutting out the middleman.

We welcome all feedback and review — whether from humans or AI — to expand this API knowledge repository and to help it converge on the most complete and AI-optimal representation of each API operation and workflow.

We're launching this repository with a significant number of initial documents:

• 2500+ OpenAPI documents from hundreds of vendors, including over 125,000 API operations, which are largely drawn from the APIs.guru repository under the terms of their Creative Commons license.
• 1000+ Arazzo workflows, which we have generated using a bespoke AI workflow agent.

We aim to rapidly build this repository to contain comprehensive information 15,000+ public APIs and to create tens of thousands of agentic tools in Arazzo format.

Join Us

We welcome all partners and collaborators who can contribute their improvements, resources or ingenuity to help build up this open-source knowledge repository on a sound footing of well-established and widely-embraced open-source standards.

Together, we can ensure a strong, open foundation for the future of agentic tool use — one that guarantees portability, interoperability, and longevity decoupled from the commercial fate of any single company. Join us in our mission to ensure that AI agents have perfect API knowledge, unencumbered by unnecessary intermediaries who charge tolls on passing traffic.