Open Intent Layer (OIL) — the open Apache-2.0 taxonomy for classifying autonomous-agent actions across 29 domains and 173 categories spanning digital enterprise operations and physical / embodied processes (manufacturing, surgical, aerial, autonomous vehicles, agriculture, energy, mining, construction, hazardous environments, logistics).
┌─────────────────────────────────────────┐
│ AGENT INTENT │
│ │
│ "deploy checkout to prod" │
│ POST /api/v2/deployments │
│ gh.actions.workflow_run.completed │
│ k8s.deployment.apps/v1.update │
│ terraform apply -auto-approve │
│ ros2:action:moveit_msgs/Pick │
│ px4:command/MAV_CMD_NAV_TAKEOFF │
│ opcua:method/siemens.tia/MoveAxis │
│ da-vinci:procedure/Suture │
│ ... │
└────────────────┬────────────────────────┘
│
│ every shape, every vendor,
│ every runtime — into one
▼ stable open vocabulary
┌──────────────────────┐
│ Open Intent Layer │
│ (OIL v2) │
│ 29 domains · 173 cats│
└──────────┬───────────┘
│
┌────────────────────┼─────────────────────┐
▼ ▼ ▼
┌───────────────┐ ┌───────────────┐ ┌───────────────┐
│ DIGITAL │ │ PHYSICAL │ │ SAFETY │
│ OIL-1XX..1400│ │ OIL-1500.. │ │ OIL-2900 │
│ │ │ 2800 │ │ │
│ deploy.prod │ │ pick.workpiece│ │ emergency-stop│
│ payments.send │ │ navigate.amr │ │ geofence │
│ access.grant │ │ takeoff.drone │ │ exclusion-zone│
│ ... │ │ surgical.cut │ │ light-curtain │
└───────────────┘ └───────────────┘ └───────────────┘
│ │ │
└────────────────────┼─────────────────────┘
▼
┌────────────────────────────────────────┐
│ Authority + Policy + Audit │
│ (any vendor, any runtime, same code) │
│ │
│ ALLOW · DENY · ESCALATE │
└────────────────────────────────────────┘
OIL provides a shared language for describing what AI agents (LLM-driven, structured-payload, robotic, drone, autonomous-vehicle) intend to do. Every action — deploying code, processing invoices, picking a workpiece, navigating to a waypoint, executing a surgical primitive — maps to a stable category in this taxonomy. That enables consistent verification, authorization, audit, and policy evaluation across any agent platform, vendor, or runtime.
Without a shared intent vocabulary:
- Every system names actions differently (
POST /api/v2/deployments,gh.actions.workflow_run.completed,ros2:action:moveit_msgs/Pick) - Authorization policies don't travel across platforms
- Audit evidence is fragmented and per-vendor
- Safety reasoning ("does this action touch a safety-rated function?") is platform-specific
OIL gives you one stable code (OIL-1502 = MoveIt manipulation primitive, OIL-2902 = geofence-enforcement safety primitive, OIL-100 family = software development) that every agent runtime, gateway, and audit chain can reference.
| Code | Domain | Categories |
|---|---|---|
| OIL-100 | Software Development | 7 |
| OIL-200 | Security Operations | 7 |
| OIL-300 | Infrastructure | 7 |
| OIL-400 | Cloud Operations | 6 |
| OIL-500 | Financial Operations | 8 |
| OIL-600 | Supply Chain | 5 |
| OIL-700 | People Operations | 6 |
| OIL-800 | Revenue Operations | 5 |
| OIL-900 | Customer Experience | 5 |
| OIL-1000 | Enterprise Operations | 6 |
| OIL-1100 | Service Delivery | 4 |
| OIL-1200 | Risk Management | 5 |
| OIL-1300 | Product Operations | 4 |
| OIL-1400 | Asset Management | 5 |
| Code | Domain | Categories |
|---|---|---|
| OIL-1500 | Manipulation | 6 |
| OIL-1600 | Locomotion & Mobility | 6 |
| OIL-1700 | Sensing & Perception | 7 |
| OIL-1800 | Actuation & Control | 5 |
| OIL-1900 | Manufacturing & Production | 7 |
| OIL-2000 | Autonomous Vehicles | 7 |
| OIL-2100 | Aerial Systems | 6 |
| OIL-2200 | Surgical & Medical Robotics | 7 |
| OIL-2300 | Agricultural Operations | 6 |
| OIL-2400 | Construction & Excavation | 6 |
| OIL-2500 | Hazardous Environments | 6 |
| OIL-2600 | Energy & Utilities | 6 |
| OIL-2700 | Mining & Resource Extraction | 5 |
| OIL-2800 | Logistics & Material Handling | 6 |
| OIL-2900 | Embodied AI Safety | 7 |
The OIL-2900 family is the load-bearing safety-primitive vocabulary. Categories include emergency-stop, geofence enforcement, exclusion-zone, light-curtain, attestation requirements, and operator-in-the-loop escalation. Other categories cite from this family via the safetyCitations field in classification results, so policies can reason about safety-relevant actions independently of vertical-specific terminology.
npm install @intended-inc/open-intent-layer
# or
pnpm add @intended-inc/open-intent-layer
# or
yarn add @intended-inc/open-intent-layerimport { oil, domains, getCategory, getDomain, getDomainForCategory } from "@intended-inc/open-intent-layer";
const cat = getCategory("OIL-1502");
// { code: "OIL-1502", name: "Pick & Place", description: "Workpiece transfer, fixture loading and unloading, bin picking, and kitting." }
const parent = getDomainForCategory("OIL-1502");
// { code: "OIL-1500", name: "Manipulation", ... }
console.log(`OIL ${oil.version}`); // "OIL 2.0"
console.log(`${domains.length} domains`); // "29 domains" ┌─ Agent emits ──────────────┐ ┌─ Classifier returns ──────┐ ┌─ Runtime uses ────────┐
│ │ │ │ │ │
│ StructuredGoal { │ │ ClassifyResult { │ │ if (failClosed) │
│ schema: "ros2:..." │ ─▶ │ oilCode: "OIL-1502" │ ─▶ │ deny() │
│ verb: "pick" │ │ confidence: 0.97 │ │ if (safetyBit) │
│ actor: { kind, id } │ │ safetyBit: true │ │ require attest() │
│ } │ │ safetyCitations: [ │ │ authorize(oilCode) │
│ │ │ "OIL-2902", │ │ emit safe-default │
│ │ │ "OIL-2904"] │ │ on deny / expiry │
│ │ │ } │ │ │
└────────────────────────────┘ └───────────────────────────┘ └───────────────────────┘
Same loop, whether the agent is an LLM emitting JSON, a ROS2 node emitting an action goal, an OPC-UA client invoking a method, or a drone autopilot issuing a MAVLink command.
OIL is the standard. Production-grade enforcement — classification, policy DSL, audit chain, signed Authority Tokens, edge verifier — is the Intended Authority Runtime, a commercial product built on top of this taxonomy. Same shape as Linux + Red Hat, OAuth + Okta, Postgres + RDS.
- Authority Runtime: https://intended.so/platform
- Physical-AI integration guide: https://intended.so/physical-ai
You can adopt OIL without buying the runtime — it's Apache-2.0 and stable across vendors.
OIL is being prepared for submission to:
- IEC TC 65 (industrial process control)
- ISO TC 299 (robotics)
- ISO/IEC JTC 1/SC 42 (artificial intelligence)
Track the work, contribute taxonomy proposals, or push back on category boundaries via GitHub Issues on this repo.
- v2.0 (current) — 29 domains, 173 categories. Adds the 15-domain physical / embodied process expansion (Manipulation, Locomotion, Surgical, Aerial, Autonomous Vehicles, Agriculture, Construction, Hazardous Environments, Energy, Mining, Logistics, Embodied AI Safety, etc.).
- v1.0 — 14 domains, 80 categories. Digital operations only. Frozen; consumers should migrate to v2.
Issues + PRs welcome. For taxonomy changes (new domains, new categories, renaming):
- Open an Issue with the proposed change + rationale
- Show at least one real-world agent integration that needs the new category
- Propose the OIL code (next available in the appropriate domain range)
- Send a PR with the taxonomy edit + a CHANGELOG entry
Apache License 2.0. See LICENSE for the full text.