AI Agent Allowlist
Home Page-Types Database Agent Guardrails 2026 Incidents API Docs Pricing
Resources
Use Cases Industries & Buyers Learn: Core Concepts Implementation Guides Comparisons Schema & Data Reference FAQ Glossary
Why It Matters
2026 Agent Incidents Category Targeting Database Refreshes Contact Customer Login
Download Free Sample
the canonical glossary · 44 terms

The Agent Guardrails Glossary, A to Z

44 terms spanning the data (page type, verified URL, IAB category), the policy model (default-deny, egress rule, zero trust), and the product mechanics (lookup, quota, database license) — each defined in a few sentences with its own anchor link, so any term here can be linked to directly from anywhere else on the site.

44Terms defined
19Letters represented, A to Z
40–80Words per definition
28Page types this glossary explains

Every term below carries its own anchor — click the # beside any definition, or add #term-name to this page's URL, to link straight to it from a spec, a ticket, or another page on this site. Several of the definitions elsewhere on the site link back here rather than re-explaining a term inline, which is the whole point of keeping one canonical, alphabetically sorted reference instead of scattering slightly different phrasings of the same idea across dozens of pages.

Terms here fall into three groups, and knowing which group a word belongs to makes it easier to find what you actually need. Data terms describe the dataset itself — page type, verified URL, IAB category, Open PageRank — the vocabulary for talking about what a lookup returns. Policy terms describe the enforcement model built around that data — default-deny, egress rule, zero trust, least privilege — the vocabulary for talking about how a decision gets made. Product terms describe how the data and policy model are actually bought and deployed — lookup, quota, database license, OEM licensing — the vocabulary for talking about integration and cost. Most terms below sit clearly in one group, though a few, like page type itself, straddle more than one on purpose, since the data and the policy built on it are not really separable in practice. If a question format suits you better than a glossary, the full FAQ covers much of the same ground as 28 questions and answers instead of standalone definitions.

ABCDEFGHILMOPQRSUVZ

A

Action type #
One of the 8 page types added in the current reclassification cycle where an agent does something rather than reads: signup, password_reset, cart, checkout, upload, post_create, comment, subscribe. Because these pages create accounts, spend money, or publish content, they are the surfaces almost every agent policy denies by default, verified the same way as any other page type: never guessed.
Agent framework #
A library or SDK — LangChain, the OpenAI Agents SDK, browser-use, Playwright-driven agents — that structures how an autonomous agent plans steps, calls tools, and browses the web. A page-type policy check plugs into whichever hook or tool-wrapper point the framework exposes, without changing the framework itself.
Agent gateway #
A proxy or API layer positioned between an agent and the open web that inspects and controls its outbound requests. Most existing gateways evaluate prompts and content; adding a page-type check is what lets the same gateway also evaluate where an agent is trying to go, not just what it is trying to say.
Allowlist #
A policy that names exactly what is permitted and denies everything else by default — the opposite of a blocklist, which names what to deny and permits everything else. An allowlist scales better against an unpredictable, ever-growing web because the burden of proof sits on the destination, not on the defender.
API key #
The credential issued the moment an API subscription is activated, used to authenticate every lookup request. Sent as either an X-API-Key HTTP header (recommended for production) or an api_key query parameter; a missing or invalid key returns a 401 response.
Audit log #
A record of every navigation decision an agent's policy engine made — the URL, resolved page type, matching rule or layer, and the allow/deny/flag result. An audit log turns a policy from a black box into something a later reviewer can reconstruct and act on after an incident.

B

Blocklist #
A policy that names what is forbidden and permits everything else by default. Blocklists require constant, reactive updates to stay ahead of new destinations, which is exactly the maintenance burden an allowlist built on a 40M-domain verified database is designed to remove.
Bulk enumeration #
Systematically requesting large portions of the dataset through the metered lookup API rather than making live, per-URL calls. This is prohibited under the API's fair-use terms; teams that need the data in bulk are expected to buy a database license instead of enumerating it through the API.

C

CASB #
Cloud Access Security Broker — a security layer that inspects and controls traffic between users and cloud services. Its core concepts (visibility, policy enforcement, and shadow-IT discovery) map cleanly onto agent web traffic once the traffic in question is an autonomous agent's own outbound requests rather than a human's.
Checkout page #
The page type where shipping, billing, and payment details are actually submitted to complete a purchase. One of the three commerce action types (alongside cart and subscribe), and almost universally denied by default, with a purchase-authorized agent under human approval as the rare, narrowly scoped exception.
Credential surface #
Any page type that authenticates a user or provisions new access — login, signup, password_reset. These carry account-takeover risk if an agent reaches them, which is why they are grouped and denied together in nearly every sample policy on this site.

D

Database license #
A one-time, on-prem purchase of the page-type database, loaded into your own infrastructure with no per-lookup network calls. Tiers run from 10M domains ($14,999) to 30M ($49,999), with optional monthly refreshes at 30% of the license price per year; 40M+ and OEM cuts are available on request.
Default-allow #
A policy posture that permits any request not explicitly named as dangerous. It requires an ever-growing denylist that is structurally always behind whatever new site, path, or redirect an agent encounters next — the posture default-deny exists to replace.
Default-deny #
A policy posture where any request the check cannot positively resolve to an explicitly allowed page type is refused rather than permitted, whether because the domain is unclassified, the page type is not on the allow list, or the lookup itself failed. This is the posture the entire product is built to support.
Domain root #
The verdict_scope value returned when a bare domain, rather than a full URL, is submitted to the lookup API. A domain-root verdict describes the domain's baseline evaluation and is not a judgment about any specific deeper page reachable from it.

E

Egress rule #
One of roughly 40 method-aware URL-pattern rules in the Egress Rules Library, matching risky shapes — wiki edits, WebDAV writes, plugin installs, signup patterns — on any domain, whether or not that domain has a row in the page-type database. Ships with every database license and is evaluated server-side by the API.
Enterprise browser #
A managed browser product that enforces organizational policy on web access for employees and, increasingly, their agents. A natural integration point for page-type rules to sit alongside existing category-based content filtering.

F

Fair use policy #
The terms governing API plans: live, per-URL lookups from your own applications and agents, with systematic bulk enumeration or replication of the underlying database prohibited. Teams with bulk needs are directed to database licenses instead.
Filtering category #
One of 59 categories in a dedicated web-filtering taxonomy built specifically for block/allow decisions — the categories firewalls, DNS filters, and secure web gateways act on. Distinct from the broader, more granular IAB content taxonomy, though both ship on every domain record.
Forward proxy #
A network intermediary that every outbound request from a client passes through. Enforcing page-type policy at a forward proxy is more robust than an SDK-level hook alone, since an agent cannot route around network-layer enforcement the way it might bypass a single tool wrapper.

G

Gateway domain #
A shared, third-party domain — a hosted Stripe Checkout session, a PayPal flow — that a merchant's own checkout redirects to for the actual payment step. It is not part of the merchant's database record; it is classified independently on its own terms, exactly as the redirect itself treats it as a separate domain.

H

High-Value Host List #
A curated list of roughly 60 dangerous or high-value hosts — cloud metadata endpoints, cloud consoles, package registries, tunnel sinks — checked ahead of the page-type database and the egress rules. A hard deny on this list wins regardless of what any other layer would otherwise decide.

I

IAB category #
A domain's classification in the Interactive Advertising Bureau's content taxonomy — 700+ categories across IAB v2 and v3, Tier 1 through 4 — describing what a site is broadly about. Included in every page-type lookup alongside the narrower web-filtering category.

L

Least privilege #
The principle that an agent should be granted only the access its specific task requires, and nothing more. An explicit, page-type-level allow list is a direct expression of least privilege for agent web browsing, in contrast to a coarse domain-level allow that grants access to every page on a site at once.
Login page #
The page type where a user authenticates. A credential surface with wide URL variation across domains — /login, /signin, an identity-provider subdomain — which is why it is verified per domain rather than guessed from a common path.
Lookup #
One API request for one domain or URL. A lookup is billed as a single unit against a plan's monthly quota regardless of how large the returned record is or how many page types, categories, or fields it contains.
Lookup API #
The per-request HTTP endpoint, GET /api/check, that returns an allow/deny verdict and the domain's page-type record. Priced from $99/month for 90,000 lookups (Pro) up to $1,997/month for 2,000,000 lookups (Business), with custom volumes beyond 10M/month.

M

MCP server #
A server implementing the Model Context Protocol, exposing tools — often including a fetch or browse capability — to an AI agent. The fetch handler inside an MCP server is a natural place to call a page-type check before honoring a browse request.

O

OEM licensing #
Custom licensing for products that redistribute this data inside their own shipping software — a secure web gateway, enterprise browser, or agent framework. Covers redistribution rights, custom delivery formats, and a refresh cadence matched to the licensee's own release cycle, priced individually.
Open PageRank #
A domain's popularity and link-authority score on a 0-to-10 scale, included in every lookup as trust context alongside global_rank. Useful for policies that weight an unranked or brand-new domain differently from an established one.

P

Page type #
A classification of what a specific URL on a domain is for — login, pricing, checkout, documentation, and 24 more — rather than what the domain as a whole is about. Up to 28 page types are verified per domain, each with its own confirmed URL.
Page-type database #
The 40M+-domain repository of verified page-type URLs this entire product is built on, covering 99.99% of active internet usage and built by analyzing 10 billion+ links. The source of truth behind both the API and the licensed database deliveries.
Password reset page #
The page type for initiating an account credential reset. A credential-surface action type, since completing the flow changes account access, and denied by default under nearly every agent policy on this site.
Policy engine #
The component — whether a tool wrapper, a framework hook, or a forward proxy — that evaluates every agent request against the four enforcement layers (host list, database, egress rules, default-deny) and returns an allow, deny, or flag decision.
Prompt injection #
Hostile content embedded in a web page, designed to redirect an agent's own reasoning toward an unintended action. A URL-level policy check enforced outside the model contains the resulting blast radius even when an injection successfully alters what the model decides to attempt.

Q

Quota #
The number of lookups included in an API plan per 30-day billing cycle, from 90,000 on the entry Pro plan up to custom enterprise volumes beyond 10 million a month. Unused lookups do not roll over; requests beyond the quota return 403 until the next cycle.

R

Rate limit #
A per-IP burst guard of roughly 240 requests per minute, applied independently of the monthly quota on every plan, to smooth traffic spikes. Hitting it returns 429 and consumes no quota.
Refresh cycle #
The optional monthly re-delivery of a database license with re-verified classifications and newly screened domains, priced at 30% of the license price per year. Without it, a database license is a one-time snapshot frozen at delivery-date accuracy.

S

Signup page #
The page type for creating a new account. Both an action type and a credential surface, since it provisions new access, and typically denied by default alongside login and password_reset.
Subscribe page #
The page type for starting a recurring-billing relationship. Distinct from a one-time checkout because it commits to an ongoing charge and often a persistent account state, so agent policies can deny it separately from a one-off purchase.

U

Upload page #
The page type for submitting files, datasets, or other content to a domain. A write surface implicated in real 2026 incidents, including the Hugging Face dataset-upload breach, and denied by default under a default-deny posture.
URL verdict #
The allow, deny, or flag decision the API returns for a specific URL and HTTP method, evaluated in one call against the host list, the page-type database, and the egress rules, falling to default-deny if none of the three produces a positive match.

V

Verified URL #
A URL confirmed live on a domain through actual link traversal and functional testing, never assumed from a path pattern. This is the core guarantee behind the page-type database: a path only appears in a record if it was seen actually working on that domain.

Z

Zero trust #
The security principle of verifying every request rather than trusting it because of its network location or apparent origin. Applied to autonomous agents, zero trust means checking every URL an agent is about to open, every time, rather than trusting a domain because it was allowed once before.
Where these terms come from

Read the full argument behind any term

Every definition above is deliberately short, 40 to 80 words, enough to know what a term means and how it is used but not enough to make the case for it. For the reasoning, the trade-offs, and worked examples behind a specific term, these are the pages that go deeper — and the 2026 incidents that made several of these terms matter in the first place, from credential surface to upload page, are covered incident by incident on the incident analysis page.

See the terms in action, in real data

Download the free 100-domain sample and match these definitions to actual verified records.

Download the Sample