Integrations and public interfaces

Available now Nitsor publishes a small set of public interfaces, and none of them is a product API. This page lists what exists today, states the boundary on each one, and shows how the documentation itself reaches an AI tool.

Available public interfaces

The current public site provides:

server-rendered marketing and customer-documentation pages;
a public contact form;
ordinary HTML links and a sitemap;
per-page Markdown documentation exports;
an llms.txt index and an llms-full.txt export;
curated, read-only context bundles for agents;
a documentation search endpoint used by the public docs interface.

These interfaces help people read about Nitsor and contact the team. They are not APIs for product data.

Contact form

The form accepts a name, email, optional company, and message. Automated tests cover validation and safe email formatting. They do not prove live email delivery, a response time, spam resistance, or a customer-support service. Do not submit secrets, credentials, sensitive scans, patient or customer data, storage URLs, or confidential strategy.

Contact form validation

Validation applies after leading and trailing whitespace is removed.

Field	Required	Exact validation
Name	Yes	Trimmed; 1 to 120 characters.
Email	Yes	Trimmed; valid email address; 254 characters or fewer.
Company	No	Trimmed; 120 characters or fewer.
Message	Yes	Trimmed; 10 to 4,000 characters.

Agent-readable documentation

Each documentation page is also served as clean Markdown at the same address with .md added. An llms.txt index lists every page with its description, and llms-full.txt combines the complete set. Both are published by the documentation host at the documentation root. Pending — agent discovery documents. The documentation host publishes machine-readable cards at /.well-known/agent-card.json and /.well-known/mcp.json, and answers a documentation MCP server at /mcp. All three were reached on the deployed origin on 24 August 2026. They stay Pending because both cards name a Mintlify-internal hostname as their own address rather than the documentation origin you are reading, so an agent that follows the address in the card leaves the public origin. Read the cards, but prefer the origin you fetched them from. Agents should cite the HTML source URL included in each export and preserve availability labels. Agent-readable documentation grants no permission. No agent can create or approve a Nitsor release today. Each written page starts as a Markdown file stored with the website code. Markdown uses plain-text formatting markers. This project uses MDX, a form of Markdown that can also include reviewed components. Mintlify is the publishing system that turns those files into the pages you are reading. Each figure on this site states its own evidence status and carries the same facts as text.

Output	What it is	Checked on a deployed origin
22 HTML pages	Every page in this documentation set.	Yes
Per-page Markdown	The same page at the same address with `.md` added.	Yes
Search	The documentation search control.	Yes
`llms.txt`	An index of every page for AI tools.	Yes
`llms-full.txt`	The complete text of every page in one file.	Yes
Agent discovery documents	Machine-readable cards under `/.well-known/`, plus a documentation MCP server at `/mcp`.	Pending

The last row is Pending: the agent discovery documents are published and reachable, but each card names a Mintlify-internal hostname as its own address rather than this documentation origin. Agent-readable outputs expose no actions and grant no permission.

The export route is part of this site and does not require a Nitsor account. The public origin is https://docs.nitsor.com, verified on 24 August 2026. For command-line use, set NITSOR_DOCS_ORIGIN to the origin of the reviewed site you are reading, without a trailing slash.

curl
Python
JavaScript

curl --fail --show-error --silent --max-time 10 "$NITSOR_DOCS_ORIGIN/index.md"

import os
import urllib.request

origin = os.environ["NITSOR_DOCS_ORIGIN"].rstrip("/")

with urllib.request.urlopen(f"{origin}/index.md", timeout=10) as page:
    print(page.read().decode())

const origin = process.env.NITSOR_DOCS_ORIGIN;

if (!origin) {
  throw new Error("Set NITSOR_DOCS_ORIGIN to the reviewed documentation origin.");
}

const page = await fetch(new URL("/index.md", origin), {
  signal: AbortSignal.timeout(10_000),
});

if (!page.ok) {
  throw new Error(`Documentation request failed with HTTP ${page.status}.`);
}

console.log(await page.text());

Each example stops after ten seconds and treats an HTTP error as a failed request instead of printing the error page as documentation. Reading these files grants no product access or permission.

Context bundles

A context bundle is a prepared set of related pages. It makes focused reading easier and grants no permission or product access. Start with the llms.txt index, then choose the smallest bundle that covers your question.

Bundle	What it covers	Pages included
Evaluate Nitsor	Current status, data boundaries, integrations, deployment, and a first action.	Introduction; Quickstart; Product status; Data boundary; Integrations and public interfaces; Deployment and security
Design-partner onboarding	Fit, prerequisites, onboarding questions, and the security evidence a partner asks for.	Quickstart; Product status; Design-partner onboarding; Deployment and security
Dataset release	The planned model for versioning, roles, review, and dataset releases.	Product status; Version model; Roles and provenance; Review, consensus, and adjudication; Release a dataset; Glossary
Programmatic access	Product status, public API, command-line interface, and data-model contracts.	Product status; Integrations and public interfaces; Public HTTP API; Command-line interface; Data model and contracts

Each bundle carries an estimated size. Those estimates are derived from the length of the text, not from any provider’s billing, so treat them as a rough guide when choosing what to load.

What the Markdown and HTML versions must agree on

Fact	Must agree	May differ
Main topic	Yes	No
Availability status	Yes	No
Prerequisites	Yes	No
Limits	Yes	No
Next steps	Yes	No
Navigation controls	No	Yes — they may appear only in HTML

If the two versions make contradictory claims, do not rely on either claim; use Product status and contact the team with the page URL and exact wording.

Programmatic and self-host surfaces

Implemented contracts do not all have the same availability. The API, CLI, and backend data contracts require an agreed design-partner environment. The self-host page is a target contract that distinguishes tested bootstrap mechanics from a supported product. None of these pages provides a public installation or sign-in path.

Surface	Availability	Current boundary	Reference
Public HTTP API	Limited design-partner access	16 implemented routes with hashed, scoped agent-credential authentication; no supported public environment yet.	Public HTTP API
Command-line interface	Limited design-partner access	Five tested command families; no published release pipeline or public installation command.	Command-line interface
Data model and contracts	Limited design-partner access	Replay-tested backend contracts; no supported operator product, and some target contracts remain planned.	Data model and contracts
Self-host mechanics	Planned — not available yet	Tested `nitsor up` bootstrap mechanics start dependencies, not a complete supported Nitsor product.	Self-host mechanics
SDK, MCP, webhook, and agent actions	Not available	No public package, server, webhook, or action surface is published.	—

Before depending on a future interface, work through the canonical pre-integration acceptance gate. Every item there — from a versioned release through export and migration behavior — applies to any interface listed as planned on this page.

Search and machine consumption limits

Search ranking is intended for this small documentation set, not as a general knowledge base. Token counts are estimates based on text size, not provider-specific billing measurements. Agent answers remain synthetic and must not be treated as product or customer validation.

Common mistakes and limits

A .md documentation URL is not a product endpoint.
A context bundle is not a tool permission.
A description of a workflow is not an SDK example.
Search visibility does not change product availability.
Generated exports must not be copied into a private operational system without checking their public status and date.

Next step

If you need a product interface, write the smallest executable acceptance test it would have to pass, then bring it to Design-partner onboarding.

Outcome	Choose only an interface that is genuinely available and record acceptance tests for future dependencies.
Availability	Available now
Audience	ML engineers, Data engineers, Software engineers, AI agents
Prerequisites	Read Product status
Last verified	2026-08-23

Get started

Concepts

Guides

Reference

Available public interfaces

Contact form

Contact form validation

Agent-readable documentation

How one written source becomes six public outputs

Context bundles

What the Markdown and HTML versions must agree on

Programmatic and self-host surfaces

Search and machine consumption limits

Common mistakes and limits

Next step

​Available public interfaces

​Contact form

​Contact form validation

​Agent-readable documentation

​How one written source becomes six public outputs

​Context bundles

​What the Markdown and HTML versions must agree on

​Programmatic and self-host surfaces

​Search and machine consumption limits

​Common mistakes and limits

​Next step

Available public interfaces

Contact form

Contact form validation

Agent-readable documentation

How one written source becomes six public outputs

Context bundles

What the Markdown and HTML versions must agree on

Programmatic and self-host surfaces

Search and machine consumption limits

Common mistakes and limits

Next step