generated: '2026-07-20'
method: searched
source: docs.pyth.network, openapi/pyth-hermes-openapi-original.json, openapi/pyth-benchmarks-openapi-original.json
authentication:
  style: >-
    Public read APIs today (Hermes / Benchmarks); Pyth Pro uses `Authorization: Bearer {key}`.
    See authentication/pyth-authentication.yml.
idempotency:
  supported: false
  reason: >-
    All Hermes and Benchmarks operations are read-only GET requests; there is no state-mutating
    write surface and therefore no idempotency-key contract. (On-chain writes happen via smart
    contracts, outside the REST API.)
pagination:
  style: offset-limit / query-parameter filtering
  notes: >-
    Benchmarks price_feeds and TradingView shim endpoints accept query params (query, asset_type,
    limit, offset on the MCP get_symbols surface). Hermes price update endpoints select feeds by
    explicit price feed id(s) rather than paging.
data_model:
  price_scale: actual_value = price * 10^expo (expo typically -8)
  confidence_interval: each price carries a `conf` field (same scale as price) for risk-adjusted pricing
  staleness: consumers must enforce a max-age (getPriceNoOlderThan / publish_time), pull model
encoding:
  formats: [json, hex, base64]
  notes: BinaryUpdate encoding selectable via `encoding` query param (hex|base64) on Hermes v2 updates.
streaming:
  server_sent_events: GET /v2/updates/price/stream (Hermes SSE)
  websocket: Pyth Pro (Lazer) low-latency streaming with configurable channels (1ms-1s)
versioning:
  scheme: uri-path
  current: v2 (Hermes), v1 (Benchmarks)
  legacy: Hermes /api/* endpoints are the v1 surface, superseded by /v2/*
rate_limiting:
  signaling: Hermes public endpoint is rate-limited; a Pyth API key will be required from 2026-08-18.
error_envelope:
  hermes: plain HTTP status codes (404 for unknown feed/timestamp)
  benchmarks: FastAPI validation envelope (HTTPValidationError, 422) + TradingView error response shape
cross_links:
  errors: errors/pyth-problem-types.yml
  lifecycle: lifecycle/pyth-lifecycle.yml
  authentication: authentication/pyth-authentication.yml