Skip to content
AIGP

RFC-011: Agentic Context Protocol (ACP) — 2. Core Principles

AIGP Specification › RFC-011: Agentic Context Protocol (ACP) › 2. Core Principles

← 1. Purpose · Section index · 3. Architecture →

2. Core Principles

  1. Context IS Agency

    • An agent’s capability is defined entirely by the context it receives.
    • No context = no action. Partial context = limited action.
    • The protocol controls what the agent can do by controlling what it knows.

  2. Distributed Assembly

    • Context is not stored in one place. It is assembled at runtime from multiple providers.
    • Each provider owns its domain of truth.

  3. Schema-First

    • Every context payload conforms to a declared schema.
    • Receivers validate incoming context against the schema before use.
    • Schema mismatches are protocol errors, not runtime surprises.

  4. Governed Fetching

    • Every context fetch is an AIGP-governed operation.
    • The governance authority (governance-server) controls which agents can fetch which context from which providers.

  5. Dual Use: Execution AND Manufacturing

    • ACP context is used for real-time execution (run a remediation, assign a role).
    • ACP context is also used for manufacturing artifacts (compile a Python script, generate a YAML config, assemble a JSON policy).
    • The protocol makes no distinction context is context.

← 1. Purpose · Section index · 3. Architecture →