Google has released the Open Knowledge Format, or OKF, an open specification designed to make it easier to package and move knowledge between people, tools, and AI systems. The company says the goal is to address a common problem in enterprise AI: useful context is often spread across too many disconnected systems for models and agents to use effectively.
The announcement, published by Google Cloud’s data analytics team, positions OKF as a vendor-neutral format for representing the curated knowledge that modern AI systems need. Google describes it as a way to formalize a familiar pattern already used by developers and data teams, where important information lives in markdown files with YAML frontmatter and links between documents.
According to Google, the first version of the specification, OKF v0.1, stores knowledge as a directory of markdown documents. Each document represents a concept such as a table, dataset, metric, runbook, or API. The files can be read by humans, indexed by search tools, stored in version control, and consumed by agents without requiring a new runtime or software development kit.
The company says the format is intentionally simple. It relies on standard markdown, ordinary files, and a small set of structured fields in YAML frontmatter. In Google’s view, that simplicity should make the format portable across organizations and software products, while avoiding the need for a new proprietary knowledge service.
Google argues that many organizations already have the information AI agents need, but it is scattered across wikis, metadata catalogs, shared drives, code comments, notebook cells, and the memories of experienced employees. When an agent needs to answer a question such as how to calculate weekly active users, it may have to gather details from several incompatible sources.
The company says this fragmentation forces every team to solve the same context-assembly problem separately. It also creates duplicated work for data catalog vendors and locks knowledge into whichever system produced it in the first place. OKF is intended to provide a shared format instead of another isolated platform.
The specification borrows from what Google and others see as an emerging "knowledge as wiki" approach. In this model, teams maintain a living library of markdown documents that can be updated by humans and AI agents alike. Google says the structure is already familiar to users of tools such as Obsidian, Notion, and Hugo, as well as to teams using agent-specific convention files and metadata-as-code workflows.
In its current form, OKF defines a bundle as a directory tree of concept files, with the file path serving as the identity of each concept. Documents can link to one another with regular markdown links, creating a graph of relationships rather than a simple folder hierarchy. The format can also include index files for navigation and log files for chronological history.
Google says the specification requires only one mandatory field for each concept, a type field, while leaving most other content decisions to the producer. The company frames this as a minimal, interoperable layer rather than a rigid data model.
The cloud provider is also releasing reference implementations alongside the specification. These include an enrichment agent that scans BigQuery datasets and drafts OKF documents for tables and views, then uses a second model pass to add documentation, citations, schemas, and join paths. Google is also offering a static HTML visualizer that renders an OKF bundle as an interactive graph in a single file.
The company said it is publishing the standard openly from the start and expects third-party producers and consumers to build around it. OKF v0.1 is meant as a starting point, Google said, with the format expected to evolve as more users test it in practice.