Google Cloud Spec Brings Order to the Chaos Surrounding AI Context

Standards and specifications launched in the recent past – most notably Model Context Protocol (MCP) and Agent2Agent Protocol (A2A) – have contributed significantly to the expansion of AI because they make it easier to operationalize the technology in enterprise use cases.

The Open Knowledge Format (OKF), just launched by Google Cloud, aims to further accelerate enterprise AI’s momentum by defining an agent- and human-friendly standard to represent the metadata, context, and knowledge that Large Language Models need to do their work. This connection between knowledge and AI has come to be referred to as an LLM-to-wiki pattern.

By standardizing such patterns, OKF will complement and build on the agent-to-back-end system specification MCP (which has hundreds of supporting vendors) and the agent-to-agent specification A2A (with over 150 supporters).

The Context Problem

Google Cloud launched OKF to address what it correctly calls a fragmented context landscape comprised of diverse, largely internal knowledge sources used by AI models: table schemas, business metrics, and so on. These sources live in fragmented systems including metadata catalogs with custom APIs, wikis in shared drives, comments within software code, and individuals’ hard drives (or possibly even their brains).

As a result of this fragmentation and lack of consistency, an AI agent that needs to answer a complex question has to build the answer from these incompatible sources, solving every problem from scratch while the knowledge remains mostly off limits outside its siloed application.

Increasingly, developers are grappling with this problem and looking to give agents a shared markdown library that increases in usability and value over time. When they’re able to do so, agents can handle the drudgery of reading and updating files, while people curate and manage content. LLMs can access multiple files with a single query – plus they don’t get bored with the painstaking, repetitive work.

The OKF Solution

While several LLM-to-wiki patterns have emerged with the goal of addressing the fragmented context problem, they don’t use common formats and therefore don’t collaborate.  What’s really needed: standardization applied to the fields that are included in every document and what those fields are called, as well as what each filename means, to name a few.

Standardization – via OKF — will ensure knowledge that’s consumed by an LLM meets these requirements:

• Anyone can produce it without an SDK
• Anyone can consume it, without performing integration
• It moves between systems, organizations, and tools without change
• It lives in version control documentation alongside the code it describes
• It’s readable by humans and parse-able by agents, with no translation layer needed

OKF includes a directory of markdown files that represent concepts that the business needs to capture including tables, datasets, metrics, and APIs. Concepts link to each other, turning the directory into a graph of relationships that is richer than the individual links. The initial specification includes conformance criteria, cross-linking rules, and a small number of reserved filenames.

OKF v0.1 represents knowledge as a directory of markdown files with YAML, or human readable, metadata placed at the beginning of a markdown file with agreed-upon conventions that let wikis from different creators be consumed by different agents.

Design Principles and Deliverables

Google Cloud said three core principles guided its OKF initiative, which is built to be

• Minimally opinionated: it requires only a type field, leaving everything else—including the content model – to the producer
• Producer/consumer independence: it separates who creates the knowledge from who consumes it. The format is standard, while the tools at each end are chosen and deployed by the user
• It’s a format, not a platform. That means it’s not tied to a cloud, database, model provider, or agent framework

Along with the spec itself, Google is shipping an enrichment agent that demonstrates one way to produce OKF, a static HTML visualizer that demonstrates one way to consume OKF, and three sample bundles that conform to OKF.

The repository, the spec, and the sample bundles are available in GitHub. Google Cloud’s Knowledge Catalog has been updated so it can ingest OKF and serve it to agents. The relevant code and examples are available here.

More AI Standards Insights:

• Google Advances A2A Protocol, Gains Microsoft and SAP Backing
• AI Agent and Copilot Podcast: Partner Opportunities With MCP
• Microsoft Showcases Real-World Use of MCP
• AI Expert Labels MCP ‘Incredibly Useful’

For a 36-Hour Immersion into the FY27 Priorities that define Partner Success in the AI Era, join us at the AI Business Solutions Partner Executive Summit, running July 22-23, 2026, in Bellevue, Washington. Register today.

---