KL4A - Turning SOPs into reviewed, evidence-linked OKF bundles for agents

[kvivek0591]

Hi Everyone,

We’ve open-sourced KL4A - Knowledge Layer For Agents (https://github.com/CogniSwitch/KL4A), an independent project that uses OKF as the foundation for turning SOPs, policies, and operational documents into knowledge that agents can inspect and use.

Our use case starts with documents written for people: PDFs, DOCX files, Markdown, and plain text. KL4A:

1. Normalizes the source documents.
2. Extracts structured, obligation-shaped claims.
3. Links each claim to its supporting source span.
4. Allows a domain expert to approve, reject, edit, defer, or comment on it.
5. Stores the knowledge, evidence, review state, relations, and decision rules as plain files.
6. Exposes the resulting bundle through a desktop app, CLI, or read-only-by-default MCP server.

The motivation is straightforward: retrieving the right document is not the same as knowing which instruction an agent received, where it came from, or whether a domain expert reviewed it. We wrote about the background story and why we released this here.

OKF is a natural foundation because its Markdown-and-frontmatter model makes the resulting knowledge portable, Git-diffable, and readable without KL4A. We are particularly interested in the v0.2 provenance, verification, lifecycle, and extensibility model.

KL4A adds a more opinionated profile for operational knowledge, including:

• Evidence spans tied to individual claims
• Persistent human-review events and rationale
• Proposed, approved, rejected, and deferred states
• Knowledge relations and decision rules
• Freshness, conflict, and validation reports
• MCP-based consumption by downstream agents

We have kept KL4A-specific fields namespaced. We are also aware that our current bundle profile is not yet a perfect drop-in implementation of every OKF v0.2 field—for example, a couple of generated-metadata field names still need alignment. We would rather be explicit about those gaps and improve the implementation with community input than claim compatibility too broadly.

Repository: https://github.com/CogniSwitch/KL4A

KL4A is currently at 0.0.1-alpha, so candid feedback is especially useful. If anyone has a real SOP or policy corpus they are comfortable testing, we would love to get feedback.

[straumat]

Nice work! on the "Consume" part, feel free to add my open source tool Kiso, it turns Open Knowledge Format (OKF) bundles into static websites for humans and AI agents to browse