Open protocol · Plain text · Local verification · v1 frozen

Context that travels with the object.

KNOBE is a plain-text format for knowledge objects that need to keep their source, history, limits, and obligations attached as they move across tools, institutions, and AI systems.

When a transcript, lesson, research note, accessibility adaptation, or AI-assisted summary leaves the system where it was created, its context can become hard to reconstruct. KNOBE helps selected context stay part of the file.

A KNOBE file can be read like ordinary Markdown. It also holds a sealed payload that records declared provenance, transformations, fidelity limits, use conditions, and an integrity hash. Verification shows whether that sealed record has changed.

View an example

When context separates from content

Knowledge objects move. The context that makes them interpretable usually doesn't.

A graduate student records an oral history. The transcript carries consent terms, a do-not-quote request, and contextual notes. A semester later, an excerpt is pasted into an AI assistant to draft an abstract. The abstract is summarized for a newsletter. A sentence from the newsletter appears in a grant report.

Illustration: as the material moves from the transcript to an AI abstract to a newsletter to a grant report, the interpretive conditions attached to it fall away. Four of four remain at the transcript, two at the AI abstract, and none at the newsletter or the grant report.

Every step was reasonable. By the fourth, the words still circulate, but the consent terms, the do-not-quote request, and the notes that governed them are gone, and no one downstream knows they ever existed. The white paper names the failure: compression without portable interpretive obligation.

A file-level gap
Software has version control

Software history is preserved through version control, commits, diffs, and inspectable change records.

Knowledge materials often lack one

Knowledge work has no equally ordinary habit for carrying the history of an idea as it moves through systems.

The AI boundary

When knowledge crosses an AI boundary, its origin, fidelity limits, and context often stop traveling reliably. The content may remain usable, while the object's interpretive conditions become harder to inspect. Transparency obligations under the EU AI Act arrive in August 2026; a sealed file gives disclosures a durable, checkable place to live.

In a KNOBE, selected context is stored inside the file, so it can travel with the work.

One file, three layers

A KNOBE is a single plain-text file. A human reads it in any editor. A machine decodes its payload. A verifier checks its payload hash.

layer 1 · yaml frontmatter
Human-scannable metadata: title, author, license, date.
layer 2 · markdown body
The document itself. No schema. No special software required to read it.
layer 3 · base64 json payload
Attribution, source context, transformation history, fidelity limits, use conditions, accessibility lineage, and a SHA-256 integrity hash.
Optional · learn by doing
New to KNOBE? Walk the Grove

A short, guided path that teaches the format by building one real sealed .knobe.md with you: no login, nothing to install. It's a friendly way to dig in and learn the idea, not the standard or fastest way to make a KNOBE. Note: Grove may take up to a minute to load on first visit.

Enter Grove →
Next steps

Understand the design

The full conceptual case, or a hands-on walk.

White paper →Grove →Studio →

Evaluate a use case

Domain guidance, plus the limits and the comparisons.

Education →Teaching →Enterprise →Government →Research →Threat model →Related work →

Implement or integrate

The normative rules, the verifiers, the corpus.

Spec →Lens →MCP server →Test vectors →Implementations →GitHub →
Every page on this site, by category
The whole picture
See how it fits together: a one-page visual overview →

The three-layer file, five domain workflows, and why context matters.