Schema
Standard fields for identity, offers, services, prices, geography, contacts, constraints and next actions.
Business data standard
Define how AI represents a business
AgntBase currently prepares the AI-readable business layer. The product direction is to define the canonical schema, normalization rules, trust layer and action routing that make business meaning harder for AI systems to distort.
Minimum standard
What “define data” requires
Not any JSON. A strict model that tells agents what a business is, what it offers, what is trusted, what is current and what actions are allowed.
Normalization rules
How to describe services, packages, prices, availability, locations, trust and edge cases consistently.
Canonical enforcement
One official source of truth so agents do not merge old bios, stale pages, competitor claims and random snippets.
Trust layer
Every field should expose source, freshness, confidence, verification status and whether owner confirmation is required.
Action model
Agents need explicit next steps: ask, compare, request quote, start checkout, book, contact or hand off to a human.
Quality gate
Profiles should pass checks before being treated as strong: no vague names, no low-signal summaries, no fake package-ready flags.
Canonical profile shape
Core object groups
This is the practical model AgntBase should keep strengthening across generated profiles and paid packages.
| Identity | company_name, domain, entity_type, canonical_html_url, canonical_json_url. |
|---|---|
| Offers | services[], products[], packages[], price_model, availability, service_area. |
| Trust | source_pages[], verified_fields[], trust_score, last_verified_at, owner_confirmed. |
| Interpretation | fit_for[], not_fit_for[], limitations[], faq[], comparison_notes[]. |
| Actions | checkout_url, booking_url, contact_form_url, messenger_url, human_review_required. |
| Protocol | llms.txt, agenthub.json, company-profile.json, optional agent-card.json when a real agent interface exists. |
Roadmap
How this evolves
The product should move in layers so we do not overpromise before the data is trustworthy.
Readable
Generate files and canonical profiles from public business signals.
Controlled
Normalize offers, services, prices, trust and allowed actions into a stricter schema.
Verified
Separate public-extracted, inferred, owner-confirmed and stale fields with quality gates.
Agent-ready
Add A2A Agent Card, safe read-only methods, checkout instructions and explicit handoff routes.
Actionable
Write capabilities only behind authentication, merchant policy and human confirmation.
Fresh
Continuously monitor sites/catalogs and update the canonical layer when facts change.
Important: AgntBase should not claim to control model behavior. The defensible claim is stronger and cleaner: we control the official business data layer that AI systems can read, verify and use for safer interpretation.