specification: API Commons Rate Limits
specificationVersion: '0.1'
generated: '2026-08-29'
method: searched
source: >-
  https://relevanceai.com/docs/admin/system-limits,
  https://relevanceai.com/docs/get-started/pricing,
  https://relevanceai.com/docs/admin/subscriptions/plans,
  https://relevanceai.com/docs/integrations/mcp/mcp-server,
  live probe of POST https://api-f1db6c.stack.tryrelevance.com/latest/agents/list
provider: Relevance AI
providerId: relevance-ai
tags:
  - AI Agents
  - Quotas
  - Concurrency
description: >-
  Relevance AI publishes NO HTTP rate limit. There is no requests-per-second or
  requests-per-minute figure anywhere in the documentation, and a live call returned no
  rate-limit header of any kind. What the provider does publish, and what actually constrains
  a caller, is a set of hard system quotas plus per-tier consumption quotas and a per-tier
  concurrency ceiling. Exceeding concurrency QUEUES work rather than rejecting it, so there is
  no 429 to back off from. This file REPLACES a 2026-05-04 scaffold that invented
  requests-per-minute figures and X-RateLimit-* headers the API does not send.
headers:
  limit: null
  remaining: null
  reset: null
  retryAfter: null
  policy: null
  observed: >-
    A live unauthenticated POST to /latest/agents/list returned no X-RateLimit-*, no
    RateLimit-*, and no Retry-After. Response headers it DID return: x-request-id, error_id,
    x-trace-id (value "x-trace-id-deprecated").
responseCodes:
  throttled: null
  quotaExceeded: null
  observed:
    422: Validation failure (missing required filter).
    403: Refused path prefix at the edge.
  note: >-
    No throttling status code is documented and none was observed. On concurrency exhaustion
    the documented behaviour is queueing, not an error.
limits:
  - id: concurrent-agent-tasks
    scope: per-organization
    metric: concurrent_tasks
    limit: null
    tier_relative:
      free: Less
      pro: Standard
      team: More
      enterprise: Custom
    window: instantaneous
    behavior_on_exhaustion: queued until capacity frees up
    detail: >-
      "The number of concurrent tasks you are allowed is determined by your subscription tier,
      NOT the type of trigger (webhook, API, or scheduled)." No absolute number is published
      for any tier, so limit is recorded as null rather than guessed. Usage is visible in
      Settings > Plan & Billing and on the Analytics page.
    source: https://relevanceai.com/docs/admin/system-limits#concurrent-tasks
  - id: actions-quota
    scope: per-organization
    metric: actions
    window: month
    limits_by_tier:
      free: 200
      pro: 2500
      team: 7000
      enterprise: custom
    behavior_on_exhaustion: >-
      Hard stop if a usage limit is configured; otherwise upgrade or buy a top-up
      ($80 per 1,000 Actions, paid plans only).
    reset: Plan Actions reset at renewal; purchased top-ups roll over.
    source: https://relevanceai.com/docs/get-started/pricing
  - id: vendor-credits-quota
    scope: per-organization
    metric: vendor_credits
    window: month
    limits_by_tier:
      free: 1000
      pro: 10000
      team: 35000
      enterprise: custom
    free_note: The Free allocation is a one-time 1,000 at signup and does not renew.
    reset: Roll over indefinitely while subscribed.
    source: https://relevanceai.com/docs/get-started/pricing
  - id: tool-run-duration
    scope: per-run
    metric: seconds
    limit: 900
    variant_limit: 86400
    detail: 'Tools: 15 minutes on the standard class, 24 hours on the long-running class.'
    source: https://relevanceai.com/docs/admin/system-limits
  - id: agent-run-duration
    scope: per-run
    metric: seconds
    limit: 900
    detail: 15 minutes for an agent after each user action.
    source: https://relevanceai.com/docs/admin/system-limits
  - id: bulk-run-duration
    scope: per-run
    metric: seconds
    limit: 86400
    detail: 24 hours for a bulk run.
    source: https://relevanceai.com/docs/admin/system-limits
  - id: mcp-sync-trigger-timeout
    scope: per-call
    surface: MCP
    metric: seconds
    limit: 120
    detail: >-
      relevance_trigger_agent times out at 120 seconds. The provider documents switching to
      relevance_trigger_agent_async + relevance_poll_agent_result (poll every 3-5s, no ceiling)
      for anything longer.
    source: https://relevanceai.com/docs/integrations/mcp/mcp-server
  - id: upload-file-size
    scope: per-upload
    metric: bytes
    limit: 104857600
    detail: 100 MB maximum upload size.
    source: https://relevanceai.com/docs/admin/system-limits
  - id: upload-csv-rows
    scope: per-upload
    metric: rows
    limit: 50000
    detail: 50K rows maximum for a CSV upload.
    source: https://relevanceai.com/docs/admin/system-limits
  - id: knowledge-raw-text-size
    scope: per-upload
    metric: bytes
    limit: 10485760
    detail: 10 MB maximum raw text size as knowledge.
    source: https://relevanceai.com/docs/admin/system-limits
limit_count: 10
customer_controls:
  - name: Usage limits
    scope: organization and project
    detail: >-
      Admins can set hard-stop limits on Credits and Actions per calendar month, with email
      notification. Available on all plans, reset at the start of each month.
    source: https://relevanceai.com/docs/admin/project-management/usage-limits
  - name: Usage alerts
    detail: Email notification only, no hard stop.
    source: https://relevanceai.com/docs/admin/project-management/usage-alerts
gaps:
  - No HTTP rate limit published and none observed.
  - No rate-limit response headers, so an agent has no runtime backpressure signal.
  - No absolute concurrent-task number for any tier.
  - No documented status code for throttling; concurrency exhaustion queues silently.