Control what AI can know and use.
CGP is the governed context layer between your data and AI models. It selects what is necessary, removes what is sensitive, enforces policy and records why every decision was made.
One control plane for context, privacy and trust.
Replace scattered prompt filters, document scrubbing and provider-specific rules with a portable contract enforced before inference.
Declare exactly what the model is allowed to receive.
Intent, token and byte budgets, freshness, provenance, privacy level, jurisdiction and retention become machine-readable constraints.
Sanitize before transmission.
Extract and transform documents locally. Unsupported or insufficiently inspected content can fail closed.
Compile the minimum sufficient context.
Rank context nodes under explicit token, byte, freshness and latency budgets instead of sending full document dumps.
Keep data from becoming instructions.
Untrusted files, web content and tool output cannot promote themselves into policies, permissions or executable actions.
Turn requirements into deterministic decisions.
CGP targets five runtime outcomes: ALLOW, SANITIZE, ROUTE_LOCAL, REQUIRE_HUMAN and DENY.
Prove what happened without exposing the original secrets.
Portable receipts capture context, transformations, routing and policy decisions for audit and debugging.
From raw source to authorized context.
Every stage has an explicit trust boundary and a documented output. Select any stage to inspect its design.
Connectivity and governance are different layers.
MCP exposes tools and resources. CGP governs the context compiled from them before that context reaches a model.
Build the open standard for governed AI context.
Use the reference implementation, contribute to the protocol or fund independent open-source development.