generated: '2026-08-04'
method: searched
source: https://docs.h2o.ai/enterprise-h2ogpte/rest-api, https://h2ogpte.genai.h2o.ai/swagger-ui/, https://genai.h2o.ai/appstore
notes: >-
  H2O.ai publishes no test-vs-live key separation, no sandbox host, and no magic test values.
  What it does publish is a live freemium tenant plus an authenticated interactive console —
  developers evaluate against real data on the shared instance rather than against a mock.
  Nothing below is invented; the absence of test fixtures is itself the finding.
test_mode:
  supported: false
  key_prefixes:
    live: sk-
    test: null
  note: >-
    All Enterprise h2oGPTe API keys carry the single sk- prefix. There is no test-mode key,
    no mode flag on the request, and no separate sandbox base URL — the only documented host
    is https://h2ogpte.genai.h2o.ai/api/v1.
console:
  url: https://h2ogpte.genai.h2o.ai/swagger-ui/
  type: Swagger UI
  live_requests: true
  authenticated: true
  note: >-
    "Browse every REST endpoint and try live requests in your browser." Requires being signed
    in to an Enterprise h2oGPTe account; requests hit the real tenant.
free_tier:
  url: https://genai.h2o.ai/appstore
  name: H2O Generative AI app store
  description: >-
    The h2oGPTe OpenAPI description states the platform is accessible through the H2O
    Generative AI app store account, "available with a freemium tier." This is the practical
    stand-in for a sandbox — a real account with real quota, not an isolated test environment.
  api_key_docs: https://docs.h2o.ai/enterprise-h2ogpte/guide/apis
key_scoping_as_blast_radius_control:
  mechanism: collection-specific API keys
  description: >-
    In place of a test mode, H2O.ai's documented containment story is key scoping: a
    Collection-specific key can only chat with the one Collection it was created against and
    cannot create, delete, or reach any other Collection or Chat. For agent development this
    is the closest available substitute for a sandbox credential.
  docs: https://docs.h2o.ai/enterprise-h2ogpte/guide/apis
test_values: []
test_clocks:
  supported: false
fixtures_and_triggers:
  supported: false
self_hosted_alternative:
  note: >-
    The open-source h2ogpt project (https://github.com/h2oai/h2ogpt) can be run locally with
    an OpenAI-compatible server, which developers use as a local substitute. It is a separate
    open-source product, not a vendor-operated sandbox for the Enterprise h2oGPTe API.
gaps:
- No test-mode API key or key prefix
- No sandbox host separate from production
- No published test fixtures, seed data or simulated failure triggers