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: Blaxel
providerId: blaxel
created: '2026-07-01'
modified: '2026-07-01'
reconciled: false
tags:
- AI
- Agents
- Infrastructure
- Sandboxes
- MCP
- Compute
- Rate Limiting
- Quotas
- Throttling
description: >-
  Blaxel constrains usage primarily through spend-tier concurrency limits rather
  than fixed per-second request quotas. Each account sits in a tier (Tier 0-9)
  that caps the number of concurrent sandboxes - roughly 10 on the free Tier 0,
  50 at Tier 1, 200 at Tier 2, and scaling to 100,000+ at Tier 9. Additional
  operational limits apply to hosted workloads: synchronous agent invocations
  run up to about 100 seconds, asynchronous invocations up to about 10 minutes,
  and batch jobs up to about 24 hours. Control plane REST endpoints may also be
  throttled to protect the platform. Specific per-endpoint request rates are not
  reconciled in this artifact.
notes: >-
  Verify current tier concurrency thresholds, control-plane request throttles,
  and workload timeouts in the Blaxel documentation and console during
  reconciliation; tier limits change as accounts add credits.
sources:
- https://blaxel.ai/pricing
- https://docs.blaxel.ai/
responseCodes:
  throttled: 429
limits:
- name: Concurrent Sandboxes
  scope: account
  metric: sandboxes
  limit: tier-based (10 on Tier 0 up to 100,000+ on Tier 9)
  notes: Primary concurrency limit; raises automatically as the account moves up spend tiers.
- name: Synchronous Agent Invocation Duration
  scope: deployment
  metric: seconds
  limit: ~100 seconds
  notes: Maximum wall-clock duration for a synchronous agent HTTP request.
- name: Asynchronous Agent Invocation Duration
  scope: deployment
  metric: seconds
  limit: ~600 seconds
  notes: Maximum duration for an asynchronous agent invocation.
- name: Batch Job Duration
  scope: job
  metric: seconds
  limit: ~86400 seconds (24 hours)
  notes: Maximum runtime for a batch job execution.
- name: Sandbox Idle TTL
  scope: sandbox
  metric: seconds
  limit: configurable (e.g. ~600 seconds default idle TTL)
  notes: Idle sandboxes are stopped after their TTL to conserve compute.
- name: Control Plane Requests
  scope: account
  metric: requests
  limit: see provider documentation
  notes: REST control plane endpoints may be throttled; honor 429 and Retry-After.
policies:
- name: Tiered Concurrency
  description: Concurrency ceilings raise as accounts add credits and move through Tier 0-9.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter and honor Retry-After on 429 responses.
- name: Idle Reclamation
  description: Idle sandboxes and hosted workloads scale to zero / stop after their TTL to limit spend.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com