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: AutoGPT
providerId: autogpt
created: '2026-05-04'
modified: '2026-08-29'
generated: '2026-08-29'
method: probed
source: >-
  Live responses from https://backend.agpt.co observed 2026-08-29, plus the 429
  and 503 response descriptions declared in
  openapi/autogpt-agent-server-openapi.json. Docs searched:
  https://agpt.co/docs/llms.txt (250 pages) has no rate-limit page.
tags:
  - AI Agents
  - AI Automation
  - Rate Limiting
  - Quotas
  - Throttling
description: >-
  AutoGPT publishes NO rate limits. This file replaces a 2026-05-04 scaffold
  whose free/professional/enterprise tiers and 10/100/1000 rpm numbers were
  invented by a bulk sweep and never existed. What AutoGPT actually meters is
  automation CREDITS, not requests: exhausting them returns 402 Payment
  Required, not 429. Three chat operations do declare a 429, and the spec's own
  prose names the three limiters behind it, but no number, window or reset is
  published anywhere and no rate-limit header is returned on any response.
limit_count: 0
limits: []
headers:
  limit: null
  remaining: null
  reset: null
  retryAfter: Retry-After
  policy: null
headers_note: >-
  No X-RateLimit-* or RateLimit-* header was returned on any live response
  observed 2026-08-29 (200, 401 and 422 all checked), and none is declared in
  either published OpenAPI. Retry-After is listed above because the spec's 503
  descriptions instruct clients to honour it — but it is not declared as a
  response header component, so a client must probe for it.
responseCodes:
  throttled: 429
  quotaExceeded: 402
  serviceUnavailable: 503
primary_meter:
  unit: automation credits
  model: >-
    Credits are charged when blocks run. Fixed for some blocks, usage-based for
    others (tokens, provider cost, elapsed time, data size, item count). For
    usage-based blocks the platform may estimate a charge before execution and
    reconcile afterwards, which can drive the balance below zero.
  exhaustion_behaviour: >-
    Paid blocks cannot start. Scheduled tasks may fail for insufficient balance.
    HTTP 402 with "Payment required: NO_TIER paywall, or insufficient credit
    balance".
  docs: https://agpt.co/docs/platform/using-the-platform/credits-and-billing.md
declared_429_operations:
  - operation: POST /api/chat/sessions/{session_id}/stream
    spec_text: >-
      "Cost rate-limit, call-frequency cap, or per-user concurrent-turn limit
      exceeded"
    limiters:
      - cost rate-limit
      - call-frequency cap
      - per-user concurrent-turn limit
    values_published: false
  - operation: POST /api/chat/sessions/{session_id}/messages/pending
    spec_text: '"Call-frequency cap exceeded"'
    limiters:
      - call-frequency cap
    values_published: false
  - operation: POST /api/chat/usage/reset
    spec_text: '"Too Many Requests (max daily resets exceeded or reset in progress)"'
    limiters:
      - max daily resets
    values_published: false
degraded_dependency_signal:
  status: 503
  spec_text: >-
    "Chat service degraded (Redis unavailable for rate limit or stream
    registry); client should honour the Retry-After header before retrying."
  note: >-
    The rate limiter is backed by Redis. When Redis is unavailable the API fails
    open to 503 rather than throttling, and Retry-After is the recovery signal.
non_rate_limits_expressed_as_409:
  note: >-
    Several caps are enforced as 409 Conflict rather than 429 — per-user skill
    limit, active-expert limit, pod limit. A client watching only for 429 will
    misread these as name collisions.
  operations:
    - POST /api/skills — "Per-user skill limit reached"
    - POST /api/experts — "Active expert limit reached"
    - POST /api/experts/pods — "Duplicate pod name, or the pod limit is reached"
probes:
  - url: https://backend.agpt.co/api/store/agents?page=1&page_size=1
    status: 200
    rate_limit_headers: none
  - url: https://backend.agpt.co/api/api-keys
    status: 401
    rate_limit_headers: none
  - url: https://backend.agpt.co/api/store/agents?page=abc
    status: 422
    rate_limit_headers: none
  - url: https://agpt.co/docs/llms.txt
    status: 200
    note: Full 250-entry docs index; contains no rate-limit or quota page.
policies:
  - name: Retry-After on degraded dependency
    description: >-
      The chat surface instructs clients to honour Retry-After before retrying a
      503. This is the only retry guidance AutoGPT publishes.
recommendation: >-
  Publish the numeric call-frequency cap and concurrent-turn limit, and emit
  RateLimit-Limit / RateLimit-Remaining / RateLimit-Reset. An agent runner that
  cannot see its remaining budget must discover the ceiling by hitting it.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com