Agents Hub · for AI agents + technical readers
Built for humans, machine-readable for agents
asxan.ai exposes structured endpoints for AI agents alongside the human evidence library. Agents can discover the corpus via llms.txt, fetch per-ingredient NC cards as JSON, or consume the long-form evidence stream via RSS / Atom feeds.
§ 1 · Machine-readable endpoints
Drop these into your retrieval pipeline · all stable URLs · versioned.
- Index
llms.txt v2 · root index
/llms.txt - Full
llms-full.txt · full corpus
/llms-full.txt - JSON
NC card JSON API · per-ingredient
/api/nc/ - OpenAPI
OpenAPI · machine-readable contract
/.well-known/openapi.yaml - Plugin
AI plugin manifest
/.well-known/ai-plugin.json - Security
Security contact
/.well-known/security.txt - Sitemap
Sitemap (XML index)
/sitemap-index.xml - Feed
RSS · evidence long-form
/blog/rss.xml - Feed
Atom · evidence long-form
/blog/atom.xml
§ 2 · Crawler policy
We welcome major AI crawlers · evidence is meant to be retrieved, cited, and reused under our dataset license.
- allowed GPTBot · OpenAI
- allowed Claude · Anthropic
- allowed Google-Extended
- allowed PerplexityBot
Full crawler policy lives in /robots.txt and /llms.txt. Dataset license at /dataset-license/.
§ 3 · Evidence methodology
How we anchor every claim to PubMed PMIDs and verbatim-cite jurisdictional authorized claims.
-
PubMed-anchored sourcing
Every dosage range, mechanism statement, and safety note in an NC card is anchored to a verified PubMed identifier (PMID). Anchors are main-session WebFetch verified · the subagent-proposed PMIDs are independently checked before they enter the data layer.
-
Verbatim regulatory claims
We surface FDA DSHEA structure-function language, EFSA Reg 432/2012 authorized claims, and ANVISA IN 28/2018 Anexo V functional claims as verbatim quotes · never paraphrased. Each claim carries its framework reference and authorization date.
-
Reverse-correction transparency
When stronger evidence overturns an earlier conclusion, we publish the correction transparently in the relevant card's safety_notes or evidence_anchor caveat · we do not silently rewrite history.
-
4-market disclaimer
Each page surfaces FDA + EFSA + ANVISA + ASXAN international positioning disclaimers in the footer · evidence summaries are educational only, not medical advice, and not evaluated by FDA / NMPA / ANVISA / EFSA for disease claims.
§ 4 · AI ecosystem surfaces
Where asxan.ai is exposed across the major AI ecosystems. Each surface is a published, machine-readable integration point so compliant retrieval, planning and chat agents can discover the evidence library without an out-of-band handshake.
- llms.txt
Anthropic Claude
llms.txt v2 indexed · /api/nc/ JSON consumable by retrieval agents
- Plugin
OpenAI ChatGPT
/.well-known/ai-plugin.json published · GPTBot allowed for browsing + retrieval
- Bot
Perplexity AI
PerplexityBot allowed · sitemap + RSS / Atom feeds for Pages discovery
- Schema
Google AI · Gemini
Google-Extended allowed · DietarySupplement + Dataset + ClaimReview structured data
- OpenAPI
OpenAPI 3.1 surface
/.well-known/openapi.yaml · contract for any compliant planning or retrieval agent
§ 5 · Agent comments (coming soon)
Future: AI agents will be able to leave structured citations and counter-evidence on any dossier · this feature is in design.