Context Fabric
Standard

The open standard for context infrastructure. 8 primitives that define how context is structured, stored, and exchanged - for humans, for AI, across any provider.

Version 0.1 Draft

What is CFS?

The Context Fabric Standard is an open specification for how context should be structured and exchanged. It defines 8 primitives that work together to create a complete context infrastructure.

Think of it like HTTP for context. HTTP doesn't care what your web server is built with - it defines how clients and servers communicate. CFS doesn't care what your context plane is built with - it defines how context is structured and exchanged.

Why a standard matters

No lock-in

Your context isn't trapped in one provider's format. Switch tools, keep context. The data model is yours.

Interoperability

Systems that implement CFS can exchange context natively. Your personal context can connect to your org's context plane.

Ecosystem

Multiple implementations, shared primitives. Competition happens on capability, not data hostage-taking.

Trust

Open spec means auditable. You can verify what a CFS implementation does with your context. No proprietary black boxes.

The 8 Primitives

Everything in CFS is built from these 8 building blocks. They're simple enough to understand, powerful enough to build anything.

Record

The atom of context

A stable identifier for any entity - person, service, document, agent. Records are the nodes in your context graph.

Link

Typed relationships

Connections between records with explicit types. "This PR closes that ticket." "This person owns that service."

View

Permissioned projections

Different consumers see different slices. Engineering sees code, sales sees customers. Same underlying data, appropriate visibility.

Tool

Safe, scoped actions

Operations agents can perform, with explicit permissions. Not "do anything" but "call this API with these constraints."

Receipt

Audit trail

Every read, every action, logged immutably. Who accessed what, when, why. Compliance without surveillance.

Policy

Machine-checkable rules

Declarative constraints that govern views and tools. "Engineers can see code, not HR data." Auditable, enforceable.

Sync

Incremental replication

Updates between context planes. Federate without centralizing. Export, import, replicate - portable by design.

Gate

Constraints that unlock

Declarative rules that gate capability. "Prove security review complete before production access." Progress through proof.

Detailed primitive documentation →

Three planes of context

CFS implementations typically organize into three planes, each with a distinct role. They can federate with each other while maintaining their sovereignty.

PCF

Personal Context Fabric

Sovereignty + Strategy

Your personal context: goals, constraints, preferences, knowledge. Encrypted, local-first, portable. Your AI self, owned by you.

OCF

Org Context Fabric

Operational Truth

The org's systems of record: repos, tickets, docs, incidents. The linked reality of what exists and what happened.

Public

Public Plane

Replication Vector

What's shared externally: docs, case studies, the standard itself. A sanitized projection for distribution and adoption.

Federation, not merging

Planes don't collapse into each other. Instead, they federate:

The standard bridges planes while respecting their boundaries.

Design philosophy

History is append-only

Corrections add new records, they don't rewrite old ones. You can always ask "what did we believe at time T?" without losing the audit trail.

Citations everywhere

Every claim traces back to source artifacts. AI summaries carry citations. Nothing is trusted without provenance.

Minimal disclosure

Consumers receive only what they need. Views filter context. Agents get capability-scoped bundles, not full access.

Verifiable correctness

Integrity hashes, merkle proofs, signed commits. You can verify that data hasn't been tampered with. Trust, but verify.

Next steps