generated: '2026-09-19'
method: searched
source: https://docs.airmdr.com/essentials/AI-Investigation-Depth-and-Usage-Limits, https://docs.airmdr.com/changelog/changelog (M28.1), openapi/airmdr-case-manager-openapi.yml
limit_count: 0
limits: []
headers: []
exhaustion_status: null
usage_allowances:
- scope: per-organization
  window: weekly (calendar week)
  limit: base weekly usage limit (value not published; set per organization)
  adaptive_threshold: 150% of the weekly limit — beyond it Darryl applies a configured investigation depth per alert type for the rest of the week; alerts are never skipped
  accumulation: unused credit carries forward up to 5x the base weekly limit
  effect: investigation depth (Level 1 Triage / Level 2 Investigation / Level 3 Deep Dive), not request rejection
  docs: https://docs.airmdr.com/essentials/AI-Investigation-Depth-and-Usage-Limits
note: >-
  No HTTP request rate limits are documented: neither spec declares a 429 response or X-RateLimit-*/RateLimit-*/
  Retry-After headers, and the docs publish no requests-per-window figure. The M28.1 release notes state that
  webhook alert ingestion applies "authentication verification, duplicate-alert handling, and rate limiting" without
  numbers (duplicates answer 409). The published limits are AI-investigation usage allowances, recorded above.