Agntbase

Research-backed articles

These pages explain how and why we build canonical company profiles, agent permissions, structured memory layer, and routing logic. Each article includes direct links to source research.

Agent-ready web

Google's UCP is a glimpse of the agent-ready web

UCP is about commerce today, but the architecture points further: agents need capability, policy, state, identity and trust layers.

Sources: Google UCP announcement, UCP.dev and Search Engine Journal analysis
Founder memo

The business record layer for the agentic web

Why businesses need a maintained AI-facing record: official facts, sources, freshness, permissions, Control Room monitoring and safe next actions.

Core Agntbase thesis: the website stays for people; Agntbase maintains the layer machines need
Control Room

What ongoing AI profile control includes

The practical difference between a $30 one-time repair and the $49/mo Control Room that keeps the AI-readable layer current.

Monitoring layer: bot reads, crawler access, failed reads, source conflicts, freshness, reputation notes, owner review and file verification
Maintenance

Why AI-readable profiles need maintenance

Why setup is only the start: freshness, profile analytics, source monitoring, owner review, safe-action checks and a simple control room.

Product layer: subscription keeps the machine-readable business profile current after setup
Source reputation

AI visibility needs a source reputation graph

Why the official source of truth needs an external narrative layer with source reputation, freshness, sentiment and risk.

New Agntbase feature: Source Reputation Graph for ranking which public sources shape the AI-readable truth layer
Machine-readable layer

llms.txt is not the product

What our hub logs and public experiments suggest about llms.txt, crawler behavior, and the maintained business context behind the file.

Core point: one file is weak; the maintained path behind it is what matters
Agent readiness

What Cloudflare's Agent Ready test gets right

Why robots, sitemap, Link headers, Markdown, Content Signals, API catalog and skills discovery point in the right direction.

Includes Agntbase position: useful checklist, but no fake endpoints for a better score
Agentic web

The missing layer between your website and AI agents

Your site can stay human-readable, but AI agents need canonical facts, source links, trust signals and next-action rules.

Explains Agntbase as the second layer beside the website, not a website replacement
Canonical profile

What is a machine-readable business profile?

A simple explanation of identity, fit, offers, evidence, freshness and safe handoff fields for AI systems.

Connects to the Agntbase business data standard and canonical profile model
AI visibility

An agent-ready website is not just SEO

SEO helps pages rank. Agent readiness helps software understand, verify and route decisions safely.

Frames the difference between ranking optimization and machine-readable business infrastructure
Trust layer

Why Agntbase does not fake MCP or OAuth signals

A higher score is tempting, but fake MCP, OAuth or API metadata makes agent discovery less trustworthy.

Explains why honest 4/6 can be better than fake 6/6 in API/Auth/MCP discovery
Product layer

Product data agents can actually use

What the agent-readability experiment suggests about offer-level facts, machine-readable catalogues and safe next-action routing.

Experiment: 1,500 noindex profiles, 5 cohorts, crawler readability separated from agent usefulness
Agent Path Map

Browser agents need safe action routes

Browser agents, WebMCP and tool protocols make websites actionable. The missing layer is a map of forms, fields, failure points and human confirmation.

Research: computer-use agents, WebMCP, NLWeb, Playwright traces, Schema.org Actions and agent safety
WebMCP / MCP

From SEO schema to WebMCP

SEO schema explains facts. WebMCP-style action discovery explains safe next steps. AI-ready websites need both layers.

Bridge: canonical profile, source links, Agent Path Map, MCP discovery and safe handoff rules
Tourism AI

How travelers plan with AI, and what tourism sites miss

Why AI trip planning needs clearer destination facts, visitor scenarios, seasonality, constraints and safe next actions.

Use case: tourism towns, destination profiles, AI-readable visitor context and local business layers
A2A status

What is already live in the Agntbase A2A layer?

A short summary of the live read-only agent layer, bot route, checkout handoff skills and the current safe boundary.

Live now: Agent Card, A2A JSON-RPC, bot hints, sitemap-agents.xml, safe read-only methods
Trust layer

Signed Agent Card is publish-ready. What does that mean?

Why publish-ready trust routes matter, what is already fixed, and what is still honestly pending before real signature activation.

Live now: signing policy, keyset route, signed artifact route, signature status method
Product fork

If browsers read websites for agents, what problem remains?

A practical fork for Agntbase: what weakens if WebMCP and browser agents make sites callable, and what still matters.

Sources: Chrome WebMCP, MCP, Forbes coverage, Google product data, and agent-led growth context
Agentic commerce

AI agents need the facts your customers see at a glance

Why online stores must translate visual intuition, taste and product context into explicit machine-readable facts.

Use case: small online stores, product descriptions, canonical profile, and AI-readable catalogue data
Agentic web

You lost the deal before sales knew

Why agentic buying workflows may compare and reject a company before any human sales conversation begins.

Sources: arXiv agentic web, manifest, structured data, and multi-agent routing references
AI visibility

AI visibility is not magic: build a machine-readable layer

Why honest AI visibility work is not a ranking promise, but a practical layer that helps machines read the business.

Sources: vc.ru note + Google AI Search guidance + arXiv agentic web references
AI analytics

GPTBot is not a lead: track AI bots server-side

Why GPTBot, ClaudeBot, OAI-SearchBot and ChatGPT-User must be measured separately from human traffic.

Sources: vc.ru note + Cloudflare/SEOmator data + OpenAI, Anthropic and Perplexity docs
Knowledge base

Practical AI readiness questions for businesses

Beauty salons, clinics, consultants and small online stores: what AI needs to understand before it can route customers correctly.

Use cases: services, trust, contacts, canonical profiles, and agent entrypoints
Technical note

Who Agntbase fits and how the technical layer works

Best fit segments, canonical entity layer, registry, protocol endpoints, events, and implementation materials.

For technical teams, agencies, partners, and deeper evaluation
Agentic web

Agentic web interface for business profiles

Why machine-facing profile interfaces outperform ad-hoc HTML scraping for agent workflows.

Sources: arXiv 2506.10953, arXiv 2507.21206
Permissions

Agent permissions and manifest layer for AI business cards

How explicit access policies reduce unsafe assumptions and improve integration reliability.

Sources: arXiv 2601.02371, arXiv 2603.18096
Linked data

Structured linked data and memory layer for company entities

Why structured entities and memory traces are more stable than plain text snapshots.

Sources: arXiv 2603.10700, arXiv 2603.07379
Multi-agent

Multi-agent orchestration and routing for business selection

How orchestration patterns apply to shortlist, fit checks, and recommendation scenarios.

Sources: arXiv 2603.18096, arXiv 2507.21206
Playbook

AI readiness playbook: website to canonical profile

A practical rollout path from raw website data to canonical profile and measurable readiness growth.

Sources: arXiv set + Google Search guidance + McKinsey AI search report