Google Cloud Introduces Open Knowledge Format

Google Cloud is launching the Open Knowledge Format (OKF), a specification that standardizes knowledge as Markdown files, making it easily portable across systems. This new format addresses the issue of fragmented knowledge that slows down AI agents by providing a minimal and interoperable structure. By representing knowledge as a directory of Markdown files with YAML frontmatter, OKF enables seamless communication between systems and agents.

The Problem of Fragmented Knowledge

Most organizations struggle with scattered knowledge across various sources, including metadata catalogs, wikis, code comments, and individual engineers' expertise. This fragmentation forces AI agents to piece together information from multiple sources, leading to inefficiencies. Google Cloud notes that every agent developer currently solves this context problem from scratch, resulting in custom-built solutions that are not designed to work with other systems.

The Open Knowledge Format (OKF) Solution

OKF v0.1 represents knowledge as a directory of Markdown files with YAML frontmatter, containing a required "type" field and optional fields like title, description, and tags. Concepts are linked through standard Markdown links, forming a knowledge graph. This minimal and portable format allows OKF bundles to be readable in any editor, rendered on GitHub, and indexed by any search tool. OKF works with any cloud provider, database, or agent framework, making it a versatile solution for standardizing knowledge.

Reference Implementations and Integration

Google Cloud is releasing several reference implementations alongside the OKF specification, including an enrichment agent that crawls BigQuery datasets and creates OKF documents for each table. A static HTML visualizer and three sample bundles for GA4 e-commerce, Stack Overflow, and Bitcoin datasets are also available. Additionally, Google Cloud has updated its Knowledge Catalog to ingest OKF and serve it to agents, with the spec and code available on GitHub.

Key Takeaways