generated: '2026-07-21'
method: searched
source: https://docs.tecton.ai/docs/reading-feature-data/reading-online-features-for-inference-using-the-http-api
summary: >-
  Cross-cutting request/response semantics for the Tecton FeatureService HTTP API and
  the Python/Java client libraries and CLI/SDK. Tecton exposes a low-latency online
  serving API for reading feature vectors, plus a declarative SDK/CLI ("feature-as-code")
  for defining and applying features.
authentication:
  style: api-key
  header: "Authorization: Tecton-key <API_KEY>"
  detail: See authentication/tecton-authentication.yml
versioning:
  api: uri-path            # HTTP API served under /api/v1/
  current_api: v1
  sdk: semver              # Python SDK / feature framework versioned (…0.8, 0.9, 1.0, 1.1, 1.2)
  current_sdk: '1.2'
  docs_are_versioned: true # docs.tecton.ai carries per-version doc trees (0.3 … 1.2 + beta)
serialization:
  request: application/json
  response: application/json
batching:
  supported: true
  detail: >-
    The HTTP API and the Python/Java clients support reading multiple feature vectors in a
    single request (get-features-batch), and wildcard reads that return multiple vectors.
idempotency:
  supported: false
  detail: >-
    Online feature reads are inherently read-only (idempotent by nature); Tecton does not
    document a client-supplied Idempotency-Key contract for write/apply operations.
error_handling:
  style: http-status
  detail: >-
    Errors surface via standard HTTP status codes (e.g. 400 invalid request, 401/403 auth,
    404 unknown feature service, 429 rate limited, 5xx server) with a JSON error body.
rate_limiting:
  detail: >-
    Online serving is provisioned per Feature Server Group with latency budgets; throughput
    limits depend on the customer's serving configuration rather than a fixed public quota.
hosting:
  model: single-tenant clusters
  base_url_shape: "https://<customer-cluster>.tecton.ai"
  detail: >-
    Each Tecton deployment is a customer-specific cluster; the HTTP API base host is the
    customer's own Tecton cluster domain (there is no single shared public API host).
cross_links:
  authentication: authentication/tecton-authentication.yml
  lifecycle: lifecycle/tecton-lifecycle.yml
  data_model: data-model/tecton-data-model.yml