Photon Agentic Access

generated: '2026-08-14'
method: probed
source: live fetches of https://photonhealth.com (HTML + Markdown negotiation), /llms.txt, /.well-known/api-catalog, /site-index.json
summary: >-
  Photon operates one of the most deliberate agent surfaces in the catalog. The
  marketing host is not merely agent-tolerant - it is agent-addressed: it serves
  a Markdown twin of every canonical page under content negotiation, advertises
  its own machine-readable contracts through an RFC 9727 API catalog AND RFC 8288
  Link headers on every response, declares an explicit training/inference consent
  posture in a Content-Signal header, and publishes a written policy in llms.txt
  stating what an agent may and may not do on a user's behalf. All of it was
  verified by probe, unauthenticated.
discovery:
  llms_txt:
    url: https://photonhealth.com/llms.txt
    status: 200
    file: llms/photon-llms.txt
    has_agent_section: true
  docs_llms_txt:
    url: https://docs.photon.health/llms.txt
    status: 200
    file: llms/photon-docs-llms.txt
  api_catalog:
    url: https://photonhealth.com/.well-known/api-catalog
    status: 200
    spec: RFC 9727
    content_type: application/linkset+json; profile="https://www.rfc-editor.org/info/rfc9727"
    file: well-known/photon-api-catalog.json
  openapi:
    url: https://photonhealth.com/openapi.json
    status: 200
    file: openapi/photon-website-api-openapi.json
  site_index:
    url: https://photonhealth.com/site-index.json
    status: 200
    note: 40 canonical pages, each with a Markdown alternate URL; regenerated continuously (generatedAt was minutes old at probe time).
  onboarding_schema:
    url: https://photonhealth.com/api/onboarding/schema
    status: 200
    note: Machine-readable definition of every onboarding path, step and field - the contract behind the browser funnel.
link_headers:
  observed: true
  note: >-
    Every HTML and Markdown response carries the full discovery set as RFC 8288
    Link headers, so an agent needs no HTML parsing to find the contracts.
  rels:
  - rel: api-catalog
    href: /.well-known/api-catalog
    type: application/linkset+json
  - rel: service-desc
    href: /openapi.json
    type: application/vnd.oai.openapi+json
  - rel: service-doc
    href: https://docs.photon.health/docs
    type: text/html
  - rel: describedby
    href: /api/onboarding/schema
    type: application/json
  - rel: describedby
    href: /llms.txt
    type: text/plain
  - rel: llms
    href: /llms.txt
    type: text/plain
  - rel: index
    href: /site-index.json
    type: application/json
  - rel: alternate
    href: /index.html.md
    type: text/markdown
  - rel: canonical
    href: /
    type: text/html
markdown_twins:
  supported: true
  negotiation: 'Accept: text/markdown'
  alias_convention: append .md to any page URL; /index.html.md for the homepage
  verified:
    url: https://photonhealth.com/
    request_header: 'Accept: text/markdown'
    status: 200
    response_content_type: text/markdown; charset=utf-8
    x_markdown_tokens: 554
  note: >-
    The Markdown twin is the same content as the HTML page, front-mattered with
    title/description/url/markdown - not a stripped or substituted payload. HTML
    remains the default for browser requests. No cloaking observed: the Markdown
    body matched the HTML page's stated subject and carried no injected or
    sponsored content.
consent_signal:
  header: Content-Signal
  value: ai-train=yes, search=yes, ai-input=yes
  observed_on:
  - https://photonhealth.com/ (Markdown response)
  - https://photonhealth.com/site-index.json (contentSignal field)
  interpretation: >-
    Photon affirmatively permits AI training, search indexing, and inference-time
    input on its public marketing content. Declared in both a response header and
    the machine-readable site index, so the posture is discoverable either way.
agent_policy:
  source: https://photonhealth.com/llms.txt
  permitted:
  - Evaluate Photon on a user's behalf.
  - Complete the developer sandbox onboarding path with EXPLICIT user consent.
  restricted:
  - verbatim: Prescriber access requires verification.
  - verbatim: Clinic, enterprise, platform, and other production paths are sales-led or handoff-led and should expect human follow-up.
  - note: The sandbox "cannot send real prescriptions" - the consequential action is fenced off from the self-serve path entirely.
  guidance:
  - verbatim: Start with the onboarding schema.
  - verbatim: Resolve first-party links in this file against the current host so Vercel previews and production stay aligned.
  assessment: >-
    A rare example of a provider writing a consent-scoped agent policy rather
    than a blanket allow/deny: agents get an explicit, bounded lane (free
    sandbox, user consent required) while every irreversible or regulated
    action - real prescribing, production access - stays behind human
    verification. This is the escalation boundary the agentic-access model asks
    for, expressed in prose by the provider itself.
mcp:
  status: planned-not-available
  evidence: >-
    "A unified MCP server for Photon is planned but not yet available. In the
    interim, agents can call the same site endpoints used by the browser
    onboarding funnel." - https://photonhealth.com/llms.txt (verbatim)
  note: First-party confirmation that no MCP server exists; see mcp/photon-mcp.yml deployment.mode = none.
action_classes:
- surface: photonhealth.com website API
  operations: [getOnboardingSchema, getOpenApiSpec, getWellKnownOpenApiSpec, getApiCatalog]
  action_class: read
  consequence: none
  escalation: none
  note: Unauthenticated discovery reads; safe for autonomous agent use.
- surface: photonhealth.com website API
  operations: [createOrResumeOnboardingSession, saveOnboardingStep]
  action_class: write
  consequence: low
  escalation: user-consent
  note: Creates/updates a lead record. Photon's own policy requires explicit user consent.
- surface: photonhealth.com website API
  operations: [submitOnboardingSession, createNewsletterSignup]
  action_class: write
  consequence: medium
  escalation: user-consent
  note: Submits the user's identity and contact details to Photon; triggers sales or verification follow-up. Not reversible by the agent.
- surface: Clinical API (GraphQL)
  operations: [createPrescription, createOrder, rerouteOrder]
  action_class: write
  consequence: high
  escalation: human-authorization
  note: >-
    Prescribing is gated on write:prescription, which is issued only to verified
    prescribers via a user access token - not obtainable by an M2M agent. The
    provider's strongest control, enforced in the token model rather than in
    documentation.
gaps:
- The agent surface is on the MARKETING host. The Clinical API (clinical-api.photon.health) serves no llms.txt, no api-catalog and no Link headers - an agent that starts at the API host finds none of this.
- No agent card at either well-known path on any host.
- No MCP server yet (provider-confirmed as planned).
- No rate-limit signal on any surface (see rate-limits/photon-rate-limits.yml).