Hormonaly — Public API & Integration Reference

The AI-native infrastructure for peptide, hormone, and longevity medicine.

This repository is the public-facing integration reference for enterprise partners and API integrators. The MCP server source is published in this repository at Hormonaly-ai/hormonaly-public. For clinical access, visit hormonaly.ai.

Table of Contents

1. Platform Overview
2. Architecture Overview
3. Partner API Reference
4. MCP Server — Tool Schema
5. Authentication
6. Rate Limits & Quotas
7. Errors & Status Codes
8. Subscription Tiers
9. Scribe API
10. Security & Compliance
11. Getting Access
12. Three-Lens Scoring
13. Webhook Events
14. API Versioning & Stability
15. Hormonaly-1 LLM
16. About Hormonaly

Platform Overview

Hormonaly combines three pillars into one platform:

• Evidence Synthesis — Multi-agent RAG over PubMed + 594 curated protocols with GRADE-style grading (A–D). 6 biomedical databases queried per response: PubMed, ClinicalTrials.gov, Cochrane, Europe PMC, OpenAlex, Semantic Scholar.
• Clinical Decision Support — Real-time answers, SOAP/DAP/Narrative notes, interaction checks, dosing, and monitoring plans grounded in verified citations.
• Workflow Tools — The operating layer for clinics, compounding pharmacies, and research teams in the peptide/hormone/longevity space.

By the numbers:

Core products available via API:

Architecture Overview

Hormonaly is built on a 42-agent orchestration architecture organized into six categories: Evidence, Research, Clinical, Content, Copilot, and Extraction — plus six continuously-running background agents.

Query Pipeline

Every Helix API call flows through the same deterministic pipeline:

Client query
│
▼
NeMo Guardrails (PII redaction · off-topic filter · nvidia/llama-3.1-nemotron-nano-8b-v1)
│
▼
Agent Router (intent classification → tier selection)
│
├── TIER_1_BEST: Claude Sonnet 4.6 → GPT-4o fallback (complex/clinical queries)
├── TIER_2_BALANCED: Claude Sonnet 4.6 → GPT-4o → Claude Haiku 4.5 fallback (Scribe, CDS, Rx)
└── TIER_3_FAST: Claude Haiku 4.5 → GPT-4o Mini → Gemini 2.5 Flash → Claude Sonnet 4.6 fallback (free tier, Three-Lens scoring)
│
▼
Multi-Database RAG Retrieval (6 databases · 80K token cap)
│
▼
Cross-Encoder Reranker (threshold ≥ 0.35, fallback 0.15)
│
▼
NeMo Parse — Full-text PMC enrichment (top 5 citations)
│
▼
Supervisor/Worker agents (complex multi-compound queries)
│
▼
Quality Gate (13 checks · 70% composite threshold · async)
│
▼
Citation Grounding (PMID verification · 4s timeout per citation)
│
▼
NeMo Output Guardrail (dosing safety scan · inline flags)
│
▼
Streaming SSE response → client

Agent Categories

Model Routing

Model routing note: All tiers use a primary + fallback chain. TIER_1: Claude Sonnet 4.6 → GPT-4o. TIER_2: Claude Sonnet 4.6 → GPT-4o → Claude Haiku 4.5. TIER_3: Claude Haiku 4.5 → GPT-4o Mini → Gemini 2.5 Flash → Claude Sonnet 4.6. NVIDIA NIM (nvidia/llama-3.3-nemotron-super-49b-v1, self-hosted on 8×H100 SXM) is used for citation grounding and the Ask Hormonaly copilot agent. Hormonaly-1 is Hormonaly's proprietary fine-tuned model — deployed on the same H100 infrastructure and progressively integrated to reduce third-party API dependence. See Hormonaly-1 LLM.

Evidence Quality Pipeline

• Reranker — Cross-encoder scores every RAG chunk ≥ 0.35 (fallback 0.15 for niche compounds)
• Token cap — 80,000 tokens max RAG context passed to LLM
• Quality gate — 13 checks across citation validity, answer completeness, clinical coherence, safety flags — composite score ≥ 70% threshold (async, post-hoc)
• Citation grounding — Every PMID verified against PubMed with 4s timeout; unverified citations are dropped
• NeMo Parse — Full PMC article text fetched for top 5 citations, structured into intro/methods/results, passed as enriched context

Partner API Reference

Base URL: https://hormonaly.ai/api/v1

Auth: Authorization: Bearer YOUR_API_KEY on every request.

Machine-readable OpenAPI spec: GET /api/v1/helix/openapi.json

All endpoint paths below are relative to the base URL.

Helix Endpoints

POST /helix/query

Submit a clinical question and receive a full Helix response with evidence grade, citations, and clinical verdict.

Request:

{
  "query": "What is the evidence for BPC-157 in tendon repair?",
  "mode": "evidence",
  "stream": true
}

Response (streaming SSE):

data: {"type":"agent_step","step":"classifying intent"}
data: {"type":"agent_step","step":"searching evidence","sources":26}
data: {"type":"agent_step","step":"verifying citations"}
data: {"type":"content","text":"BPC-157 has demonstrated..."}
data: {"type":"metadata","evidence_grade":"C","confidence":62,"verdict":"CONSIDER","citations":5,"sources_consulted":26}
data: {"type":"done"}

Response metadata fields:

GET /helix/protocols/:compound

Look up evidence-quality-rated protocols for a specific compound.

Example: GET /helix/protocols/bpc-157

Response:

{
  "compound": "bpc-157",
  "evidence_grade": "C",
  "protocols": [
    {
      "indication": "Tendon repair",
      "dose_range": "200–500 mcg/day",
      "route": ["subcutaneous", "intramuscular"],
      "cycle_duration": "4–12 weeks",
      "monitoring": ["liver enzymes at baseline", "symptom review at 4 weeks"]
    }
  ],
  "interactions": [],
  "citations": 5
}

POST /helix/dossier

Generate a comprehensive evidence dossier for a compound (async job). Advanced or Enterprise tier required.

Request:

{ "compound": "semaglutide", "sections": ["mechanism", "efficacy", "safety", "dosing"] }

Response: { "job_id": "dossier_abc123", "status": "queued" }

Poll with GET /helix/dossier/:job_id until status: "complete".

POST /helix/compare