Permission rules for AI agents. One call before they act.

Check what a domain allows before your agent acts. Get back allowed, denied, or escalate — fast, deterministic, no model in the loop.

pip install openterms-py

import openterms
result = openterms.check("github.com", "read_content")
# → { "result": { "value": true, "confidence": "high" }, "permission": "allowed" }

GET /api/v1/check?domain=github.com&action=read_content

{
  "result": { "value": true, "confidence": "high" },
  "permission": "allowed"
}

Install the SDK Read the agent guide →

511 services indexed. Open standard, free SDK, no lock-in.

The seven permissions

read_content
Fetching and parsing publicly accessible text or data from a page or document.
scrape_data
Extracting structured data from a page in bulk — beyond what a human copying by hand could do in a single session.
api_access
Calling a service's API — authenticated or unauthenticated — in a way the service intended for programmatic use.
create_account
Registering a new account or identity on the service programmatically.
make_purchases
Initiating or completing a payment transaction through the service.
post_content
Submitting, publishing, or editing content — comments, posts, reviews, or any other user-generated submission.
allow_training
Using the service's data or outputs as training input for a machine learning model.

The decision model

Check result	What it means	Agent action
allowed	The service permits this action under the stated conditions.	Proceed.
denied	The service explicitly forbids this action.	Block, log, and escalate to the human for review.
not_specified	The domain has not published an openterms.json, or this action is not declared.	Fail-closed: treat as denied unless your policy says otherwise.

Fail-closed rule: If a domain hasn't published, or the answer is unknown, treat it as not-permitted unless your policy says otherwise. The cost of a false positive (blocking a permissible action) is lower than the cost of a false negative (permitting a forbidden one).

Publish your own permissions in 30 seconds

Enter your domain. Generate reads your Terms of Service and Privacy Policy and produces a ready-to-deploy openterms.json — validated against the schema, ready to drop at /.well-known/openterms.json.

Generate yours →

Every agent that checks makes publishing worth more. Every domain that publishes makes checking worth more.

The registry

511 services already indexed — browse what's known, with confidence stated plainly.

OTA-verified entries

Production

Suitable for production agent deployments — alongside entries validated by the service owner or your own organization.

458

Machine-generated entries

Discovery & test

Useful for discovery, testing, and workflow design. Validate before relying on them in production agent deployments.

OTA = Open Terms Archive, an independent third-party record of legal documents. OTA-verified means OpenTerms-reviewed against public source material — not an endorsement, approval, or certification by the service owner.

Browse the registry →

SDKs and integrations

The Python SDK is the fastest path to integration. LangChain and CrewAI adapters handle the agent framework layer. Bulk download is available for offline or on-premise use.

pip install openterms-py

openterms-py LangChain adapter CrewAI adapter AutoGen — available browser-use — available

Bulk registry download available for offline indexing and on-premise use. Browse the registry →

Validate a file →

The format is open, the SDK is free, the lookup is one call.

Install the SDK When you need a record of what your agents did → Observe