generated: '2026-08-24'
method: searched
source: >-
  https://www.agentsea.dev/docs/multi-tenancy/, https://www.agentsea.dev/docs/guardrails/,
  https://www.agentsea.dev/api/
limit_count: 0
note: >-
  AgentSea publishes no rate limits because it operates no API - both HTTP surfaces run inside the
  consumer's own deployment, so there is no vendor-side quota to state and no 429 to document. That
  is an honest zero, not an omission.
  What the framework does provide is rate limiting as a CAPABILITY the self-hoster configures:
  a per-tenant rateLimit setting (requests per minute) on TenantSettings, quota recording via
  TenantManager.recordQuotaUsage() for the api_calls and tokens resources over hourly and daily
  periods, and TokenBudgetGuard (maxTokensPerRequest) / CostGuard (maxCostPerRequest) in the
  guardrails engine. None of these define a response-header contract.
response_headers:
  documented: false
  x_ratelimit: false
  ratelimit: false
  retry_after: false
  note: >-
    No X-RateLimit-*, no RFC 9238-style RateLimit-* and no Retry-After behaviour is documented on
    either surface, and no status code is named for exhaustion. An agent calling a self-hosted
    AgentSea deployment gets no runtime rate-limit signal from the framework itself.
exhaustion_status: null
limits: []
operator_configurable:
- name: tenant rateLimit
  scope: per-tenant
  window: minute
  unit: requests
  value: operator-defined
  docs: https://www.agentsea.dev/docs/multi-tenancy/
- name: quota - api_calls
  scope: per-tenant
  window: [hourly, daily]
  unit: calls
  value: operator-defined
  docs: https://www.agentsea.dev/docs/multi-tenancy/
- name: quota - tokens
  scope: per-tenant
  window: [hourly, daily]
  unit: tokens
  value: operator-defined
  docs: https://www.agentsea.dev/docs/multi-tenancy/
- name: TokenBudgetGuard
  scope: per-request
  unit: tokens
  value: maxTokensPerRequest, operator-defined
  docs: https://www.agentsea.dev/docs/guardrails/
- name: CostGuard
  scope: per-request
  unit: currency
  value: maxCostPerRequest, operator-defined
  docs: https://www.agentsea.dev/docs/guardrails/
upstream_note: >-
  The rate limits that actually bind an AgentSea deployment are the ones its configured LLM
  providers impose (Anthropic, OpenAI, Google and so on). The gateway's circuit breaker, failover
  and retry logic exist to absorb them.