Between humans and AI, there's a gap. A layer that should exist but doesn't. We call it the context plane — and building it is the most important infrastructure work of the next decade.
Think about what happens when you talk to an AI today.
You type a question. The model receives your words, some system instructions, maybe some retrieved context. It generates a response. The conversation either ends or continues — but fundamentally, the AI doesn't know anything about the world. It has no persistent understanding of who you are, what you're working on, or what you've discussed before.
This is by design. Large language models are stateless functions. They transform input into output without maintaining state. Every interaction is, architecturally, a fresh start.
And yet — we want AI to be useful in persistent contexts. In organizations with history. In lives with continuity. In work that spans sessions.
The gap between "stateless function" and "useful collaborator" is where context infrastructure lives.
Every technology stack has layers. Context is the layer between application and intelligence.
Without a context layer, every application builds its own ad-hoc solution: RAG pipelines, custom memory, bespoke integrations. The context plane is the missing standard.
With proper context infrastructure, AI becomes genuinely useful in real-world settings.
Your AI actually knows you. Your preferences, your projects, your history — authored by you, owned by you, portable across providers. No more repeating yourself to every new tool.
Your org becomes queryable. "Who owns this service?" "What changed before this incident?" "How does this system work?" — answered with citations, in seconds, with audit trails.
Build on a standard instead of rolling your own. 8 primitives that compose. Interoperability by default. Focus on your application, not your plumbing.
Context portability prevents lock-in. Open standards enable competition on capability, not capture. Users maintain agency in their AI relationships.
CFS defines eight primitives for context infrastructure. They're not complex — they're the obvious building blocks once you see them. The question isn't whether these primitives are right, but whether they'll be open or proprietary.
Records Stable identities for context atoms Links Typed relationships between records Views Permissioned projections Tools Scoped, audited actions Receipts Audit trail for every access Policies Declarative access rules Sync Incremental updates Gates Constraints that unlock capability We're building context infrastructure at every scale — from personal to planetary.
Own your AI self. A local-first, encrypted vault of your context. Connect it to any AI. Take it with you when you leave.
Explore Personal →Make your org addressable. Service maps, knowledge graphs, killer queries with citations and receipts.
Explore Enterprise →The open standard. Reference implementations, conformance tests, community governance. The TCP/IP for context.
Read the Spec →Context infrastructure is being built now. In a few years, the standards will be set. We're building the open foundation.