NotarIA's public surface, for AI agents.
NotarIA publishes its public content in machine-readable formats: every page is available in Markdown, and the site is described by llms.txt, a sitemap and an API catalog. There is, however, no programmatic API, and no matter data is reachable: everything concerning a notarial practice sits behind human authentication.
What is served
All of these resources are public, read-only, unauthenticated and require no API key.
| Resource | Format | Content |
|---|---|---|
| /llms.txt | text/plain | Citable product summary, main links and verifiable facts. |
| /llms-full.txt | text/plain | Detailed version: features, security, pricing, glossary, FAQ, article index. |
| /md/{path} | text/markdown | Markdown twin of every public page and article, with front matter. |
| /sitemap.xml | application/xml | Every public URL with its language alternates. |
| /robots.txt | text/plain | Per-crawler permissions and Content-Signal directives. |
| .well-known/api-catalog | application/linkset+json | Catalog of this surface (RFC 9727). |
| /openapi.json | application/vnd.oai.openapi+json | OpenAPI 3.1 description of the resources above. |
| .well-known/agent-skills/index.json | application/json | Published skills, each with the SHA-256 digest of its artifact. |
| /api/health | application/json | Service liveness probe. No data. |
Getting a page in Markdown
Two routes lead to the same document. The first is content negotiation: request the page's canonical URL with the header « Accept: text/markdown » and the response comes back as Markdown instead of HTML. The second is the explicit twin: prefix the page path with /md. So the pricing page can be read at /pricing with the header, or directly at /md/en/pricing.
Every document opens with front matter carrying the title, the description, the canonical HTML URL, the language, the last-modified date and a fidelity field. The value « exact » means the Markdown is a transposition of the same source as the HTML served; the value « summary » flags a structured summary — this is the case for the French home page, whose layout lives in interface components rather than in the content dictionary. We would rather say so than let you guess.
The response also carries two useful headers: « Link », pointing at the canonical HTML URL to cite, and « x-markdown-tokens », an estimate of the token count (roughly four characters per token) to size a context window. HTML remains the default: without an explicit Accept header, a browser always receives the normal page.
Crawl policy and Content Signals
AI crawlers allowed by name
The main search and AI crawlers are allowed explicitly in /robots.txt, which is authoritative. Being read and cited is part of our visibility strategy.
Three declared signals
Every block in /robots.txt carries a Content-Signal directive covering training (ai-train), indexing (search) and citation by a generative engine (ai-input). All three are currently allowed for public content.
Private content is out of scope
No crawler reaches a practice's data: it sits behind authentication and is isolated per practice at the database level. There is nothing to disallow in robots.txt, because there is nothing there to reach.
What does not exist
As a matter of principle, we do not publish metadata describing a capability we do not have: an agent following an inaccurate card breaks, and trust does not repair itself.
No programmatic API
There is no API key, no third-party access token, and no endpoint to read or write matters. The application runs on session-bound server actions, outside this public surface.
No OAuth endpoint
NotarIA is an OAuth client of Microsoft Entra ID, not an authorization server: it issues no tokens to third parties. We therefore publish neither authorization-server metadata, nor protected-resource metadata, nor an agent registration procedure.
No MCP server, no A2A agent
No MCP server card and no A2A agent card are published, for lack of a corresponding endpoint. Should that change, the card will follow the capability — never the other way round.
No client data
This surface serves marketing and educational content only. No matter, no party, no act and no practice data passes through it.
How to cite NotarIA
Cite the HTML URL
The canonical resource is the HTML page, given by the url field of the front matter and by the Link header of the Markdown response. The /md twins are not meant for indexing.
Date the citation
The updated field of each document gives the content's last-modified date. Québec law evolves: an undated citation ages badly.
Reproduce the disclaimer
Every document ends with the same notice: NotarIA is an assistant, it is not legal advice, and the notary reviews, signs and decides on every act. That reservation is part of the content.
Frequently asked questions
Can I access a practice's matters through an API?
No. No API exposes a practice's data, and no API key is issued. That data is available only to authorized people of the practice, after authentication, and is isolated per practice at the database level.
How do I get a page in Markdown?
Request the page URL with the header « Accept: text/markdown », or request the /md twin directly followed by the page path (for example /md/en/faq). HTML is still served by default.
May I train a model on this content?
Public content currently allows training, indexing and citation: all three Content-Signal values in /robots.txt are « yes ». /robots.txt is authoritative. Private content is never reachable by a crawler.
Does NotarIA expose an MCP server or an A2A agent?
Not today. We would rather publish nothing than advertise an endpoint that does not exist. If we open one, it will be described here and in the API catalog.
Where are these resources hosted?
In Canada. The site and the application are hosted in a Canadian region (Montréal), in line with the Law 25 design described on the Security page.
Have an integration in mind?
Let's talk with ATOM Solutions, the publisher of NotarIA.
Assistant — not legal advice. The notary reviews, signs and decides on every act.