Agentic AI Wiki

Open Knowledge Format

Properties1
tagsagents, knowledge, interoperability, protocols

2 min read

Open Knowledge Format

Open Knowledge Format (OKF) is an open, file-based format for exchanging curated knowledge between humans, tools, and agents. Google describes OKF v0.1 as a formalization of the LLM Wiki pattern using Markdown documents, YAML frontmatter, paths, and links. OKF source record · Karpathy LLM Wiki source record

Bundle model

The launch overview describes an OKF bundle as a directory in which:

  • one Markdown file represents one concept;
  • the file path identifies that concept;
  • YAML frontmatter carries structured metadata;
  • Markdown links form a graph;
  • index.md and log.md may support navigation and history.

The overview says a concept’s type is the only required field in the minimal interoperability surface, while naming title, description, resource, tags, and timestamp as useful structured fields.

Design principles

OKF is presented around three principles:

  1. Minimal opinion: standardize exchange without imposing a domain ontology.
  2. Producer/consumer independence: allow different humans, pipelines, or agents to write and read bundles.
  3. Format, not platform: avoid requiring a proprietary account, runtime, or SDK.

These principles are goals. Independent implementations and conformance evidence are needed to establish practical interoperability. OKF source record

Relationship to agent context

OKF standardizes a representation of maintained knowledge, not how an agent selects, retrieves, trusts, or applies it. An agent harness still needs search, context assembly, access control, provenance checks, and evaluation.

Analysis: A portable knowledge bundle can reduce integration friction while also carrying stale, malicious, or incompatible content. Producers and consumers need trust, version, and namespace policies.

Versioning and conformance

The announcement labels v0.1 a starting point. Claims about required fields, reserved files, link behavior, or backward compatibility should be tied to a specific specification version.

This Agentic AI Wiki is influenced by the same file-based pattern but is not claimed to be OKF-conformant. Its Obsidian-style wikilinks, frontmatter, source-record model, and publication fields require comparison with the versioned conformance rules before making that claim.

See LLM Wiki Pattern vs OKF.

Evidence status

Confidence is high for the v0.1 design intent described by its publisher and low-to-medium for universal portability or vendor neutrality in practice. The launch evidence consists of Google proofs of concept and sample bundles, not broad independent adoption.

Graph View

Backlinks