Appearance

Introduction ​

bAInquet turns any website or AI-managed project into a verified, structured, AI-optimized Knowledge Node that LLMs, agents, and MCP clients read directly instead of scraping HTML.

The problem ​

AI systems consume the web by scraping rendered HTML. That is lossy and adversarial: the model guesses at structure, parses presentation markup, and has no way to tell an authoritative statement from a stale cached page or an injected instruction. The site owner, the one party that actually knows the canonical price, the current stock, the correct hours, has no channel to hand that data to a model in a form the model can trust.

bAInquet is that channel. The owner pushes structured content in over a signed connection; bAInquet normalizes it into a typed knowledge graph, tracks where every value came from, and publishes a static, versioned set of files that an AI consumer reads as a primary source. The data is owner-supplied and the domain ownership is cryptographically verified, so a consumer can weight the node by how it was proven rather than guessing.

The three-step model ​

  push in                build                   read
+-----------+        +--------------+        +-------------+
| connector | -----> | knowledge    | -----> | Knowledge   |
| (HMAC-    |  HTTP  | graph        | export | Node files  |
|  signed)  |        | (PostgreSQL) |        | (CDN / MCP) |
+-----------+        +--------------+        +-------------+
  1. Push in. A connector sends content items to the ingestion API. Every request is HMAC-signed with a per-website key, so the platform can prove the content came from the verified owner. There are 13 prebuilt connectors, an official SDK, and a no-connector AI-agent fallback.
  2. Build the graph. A worker pipeline normalizes each item into typed entities, facts, relationships, retrieval chunks, and sources. Extraction is deterministic by default; every value records how it was extracted and how much to trust it. The graph deduplicates and merges across all of a site's content.
  3. Read the node. Exporter modules publish the graph as static files (manifest.json, ai.json, llms.txt, NDJSON record files, and more) to a CDN. AI agents and MCP clients fetch the node directly. The internal graph is the durable asset; output formats are just exporter modules.

Who it is for ​

  • Site and store owners who want AI assistants to quote correct, current facts about their products, services, and content.
  • Connector and integration authors building a push from a CMS, commerce platform, or framework that does not yet have a prebuilt connector.
  • AI / agent developers who want a verified, structured source for a domain instead of scraping, available over plain HTTP or through the MCP read server.

What makes a node trustworthy ​

  • Verified ownership. A website must prove domain control (DNS TXT, a well-known file, a meta tag, or a signed plugin) before its node is trusted. The proof method sets a trust tier that is published with the node.
  • Provenance on every value. Each fact records the extractor that produced it (cms_field, schema_org, seo_meta, text_extraction) and a confidence score. The free tier uses zero LLM work and is fully reproducible. Paid tiers add AI enrichment, and every enriched value is tagged ai_enhanced so consumers can separate it from deterministic data.
  • Signed ingestion. Content can only be pushed by a connector token bound to one verified website, signing each request with a key derived from its secret.

Next steps ​

Owner-controlled structured data for AI.

© bAInquet