Agent Intent Protocol (AIP)

Minimal semantic interaction objects for agent runtimes.

Agent Intent Protocol (AIP) is a draft semantic interaction layer for the Digital Biosphere Architecture ecosystem. It defines machine-readable intent, action, and result objects that can pass between persona-oriented systems, governance checkpoints, runtimes, and audit layers.

AIP is not a transport protocol and not a full runtime framework. Its role is narrower: provide a compact object model for what an agent is trying to do, what action is proposed or taken, and what result is returned.

Architecture Position

AIP contributes the Interaction Layer in the Digital Biosphere Architecture ecosystem.

• POP handles persona identity.
• AIP handles task intent, action, and result exchange.
• Token Governor handles runtime constraints and policy.
• MVK handles execution integrity.
• ARO-Audit handles evidence and receipts.

flowchart LR
    Persona["Persona"] --> Intent["Intent Object"]
    Intent --> Governance["Governance Check"]
    Governance --> Execution["Execution"]
    Execution --> Evidence["Evidence"]

Loading

Non-goals

• AIP is not a chat format.
• AIP is not a transport layer.
• AIP is not a full agent orchestration framework.
• AIP is not an audit record format.
• AIP is not a permission or identity substitute.

Core Objects

Intent Object

An agent_intent object declares what the actor wants to achieve and under what constraints.

{
  "schema_version": "0.1.0-draft",
  "object_type": "agent_intent",
  "intent": {
    "summary": "Find round-trip flight options from Shanghai to Singapore for next week"
  },
  "actor_ref": "pop://personas/travel-assistant",
  "constraints": {
    "max_budget_usd": 900,
    "approval_required": false
  }
}

Action Object

An agent_action object declares a specific operation proposed or executed by the actor.

{
  "schema_version": "0.1.0-draft",
  "object_type": "agent_action",
  "action": {
    "name": "call_search_tool",
    "summary": "Query the local flight-search adapter"
  },
  "actor_ref": "pop://personas/travel-assistant",
  "execution_mode": "proposal"
}

Result Object

An agent_result object declares the status and references to outputs or evidence produced by the run.

{
  "schema_version": "0.1.0-draft",
  "object_type": "agent_result",
  "status": "completed",
  "actor_ref": "pop://personas/travel-assistant",
  "correlation_id": "trip-search-001"
}

Repository Layout

• spec/ contains the draft protocol text.
• schema/ contains JSON Schema drafts for the three core object types.
• examples/ contains worked examples.
• conformance/ contains valid and invalid fixtures.
• scripts/validate_examples.py validates the examples and fixtures.
• tests/ contains a minimal pytest surface.

Conformance

Run the lightweight validator:

python3 scripts/validate_examples.py

Run the minimal test surface:

pytest

Status

• Working draft
• Semantic interaction layer only
• Intended to compose with POP, Token Governor, MVK, and ARO-Audit