apiCommonsRateLimits: 0.1
provider:
  id: runloop-ai
  name: Runloop
  website: https://runloop.ai
  documentation: https://docs.runloop.ai
  modified: 2026-05-25

scope:
  description: >
    Runloop enforces account-scoped resource quotas during the Pro Trial and exposes
    plan-tier resource ceilings through the platform. The REST API itself does not
    publish per-endpoint HTTP rate limits in the OpenAPI spec; clients should
    implement exponential backoff on 429 responses. Concurrency and resource
    counts are the effective limit surface.

quotas:
  - id: trial-running-devboxes
    plan: trial
    resource: running Devboxes
    limit: 3
    period: concurrent

  - id: trial-blueprints
    plan: trial
    resource: Blueprints
    limit: 5
    period: account

  - id: trial-snapshots
    plan: trial
    resource: Snapshots
    limit: 10
    period: account

  - id: trial-objects
    plan: trial
    resource: Storage Objects
    limit: 3
    period: account

  - id: basic-storage
    plan: basic
    resource: Free Storage
    limit: 100
    unit: GB
    period: account

  - id: pro-storage
    plan: pro
    resource: Free Storage
    limit: 1024
    unit: GB
    period: account

  - id: parallel-sandboxes
    plan: pro
    resource: Parallel Devboxes (marketed maximum)
    limit: 10000
    period: concurrent
    description: Marketing claim — "Run 10k+ parallel sandboxes". Effective limit depends on plan and approval.

retries:
  recommendation:
    backoff: exponential
    initialDelayMs: 500
    maxDelayMs: 30000
    maxAttempts: 5
    onStatuses:
      - 429
      - 502
      - 503
      - 504

notes:
  - Devbox lifecycle endpoints (create, snapshot, suspend) are heavier than read endpoints; expect lower effective throughput.
  - PTY sessions enforce a single-attach contract at WebSocket upgrade time; a second concurrent attach is rejected.
  - Suspended Devboxes do not consume compute quotas — only storage.