specification: API Commons Rate Limits
specificationVersion: '0.1'
provider: PydanticAI
providerId: pydantic-ai
created: '2026-06-12'
modified: '2026-06-12'
note: >
  PydanticAI is a client-side open-source framework that routes requests to
  upstream model providers (OpenAI, Anthropic, Google, etc.). PydanticAI itself
  does not impose rate limits — rate limiting is enforced by the chosen model
  provider. When using Pydantic Logfire for observability, ingestion rate limits
  apply to span/log/metric records as described below. The Pydantic AI Gateway
  proxies model provider calls and may enforce gateway-level limits in addition
  to upstream provider limits.
headers:
  retryAfter: Retry-After
responseCodes:
  throttled: 429
  unauthorized: 401
  forbidden: 403
limits:
  - name: Logfire Personal Plan Monthly Records
    scope: account
    metric: logs_spans_metrics
    limit: 10000000
    timeFrame: month
    plan: Personal
    note: Hard cap; upgrade to Team or Growth to get overage billing
  - name: Logfire Team Plan Included Records
    scope: account
    metric: logs_spans_metrics
    limit: 10000000
    timeFrame: month
    plan: Team
    note: Additional records billed at $2/M; configurable spend cap available
  - name: Logfire Growth Plan Included Records
    scope: account
    metric: logs_spans_metrics
    limit: 10000000
    timeFrame: month
    plan: Growth
    note: Additional records billed at $2/M
  - name: Logfire Personal AI Gateway Credentials
    scope: account
    metric: ai_gateway_provider_credentials
    limit: 3
    timeFrame: static
    plan: Personal
    note: BYOK credentials; built-in providers at 5% markup
  - name: Logfire Growth AI Gateway Credentials
    scope: account
    metric: ai_gateway_provider_credentials
    limit: -1
    timeFrame: static
    plan: Growth
    note: Unlimited BYOK at 0% markup; built-in providers at 3% markup
  - name: Upstream Model Provider Rate Limits
    scope: request
    metric: requests
    limit: -1
    timeFrame: varies
    note: >
      Actual per-request rate limits are imposed by the upstream model provider
      (e.g., OpenAI, Anthropic, Google). PydanticAI propagates these errors and
      will return a 429-equivalent exception. Use exponential backoff and
      configure retries at the Agent level with model_settings or custom retry
      logic.
retryGuidance:
  strategy: exponential_backoff
  initialDelayMs: 1000
  maxDelayMs: 60000
  maxAttempts: 5
  note: >
    PydanticAI's Agent loop has built-in retry support for tool call failures
    and model validation retries. For Logfire API 429 responses, honor the
    Retry-After header and implement client-side back-pressure before retrying
    span ingestion.