specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Union.ai
providerId: unionai
created: '2026-06-20'
modified: '2026-06-20'
reconciled: false
tags:
- AI
- ML
- Orchestration
- Workflows
- Serverless
- Rate Limiting
- Quotas
- Throttling
description: >-
  The Union / Flyte control plane (FlyteAdmin) governs work primarily through
  concurrency limits rather than published per-second API rate limits. The most
  visible quota is concurrent actions (the number of task actions a tenant can
  run simultaneously) - the Team plan documents a ceiling on the order of 1,000
  concurrent actions, with Enterprise plans negotiated far higher (50,000+).
  FlyteAdmin also applies pagination limits/tokens on its list endpoints and
  standard gRPC/HTTP error handling. Specific per-endpoint request rates are
  not published and are not reconciled in this artifact.
notes: >-
  Verify concurrent-action ceilings, list pagination limits, and any gateway
  request-rate throttling against the Union.ai pricing page and FlyteAdmin
  deployment configuration at reconciliation. Self-hosted open-source Flyte
  limits are operator-configured.
sources:
- https://www.union.ai/pricing
- https://www.union.ai/docs/flyte/api-reference/
- https://www.union.ai/docs/flyte/architecture/component-architecture/
responseCodes:
  throttled: 429
limits:
- name: Concurrent Actions
  scope: tenant
  metric: actions
  limit: ~1,000 (Team); 50,000+ negotiated (Enterprise)
  notes: Maximum task actions running simultaneously; primary throughput governor.
- name: List Pagination Limit
  scope: request
  metric: results
  limit: see provider documentation
  notes: List endpoints accept a `limit` and return an opaque `token` for the next page.
- name: Data Retention
  scope: tenant
  metric: days
  limit: 30 (Team) / 365 (Enterprise)
  notes: Execution and metadata retention window, not a request rate but a stored-data quota.
- name: Clusters
  scope: tenant
  metric: clusters
  limit: 1 (Team) / 3+ (Enterprise)
  notes: Number of compute clusters attached to the control plane.
policies:
- name: Concurrency Governing
  description: Throughput is shaped by concurrent-action ceilings per plan rather than fixed RPM/TPM caps.
- name: Pagination
  description: Use the returned token and a sane limit when listing projects, workflows, tasks, launch plans, and executions.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter on 429/503 and honor Retry-After when present.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com