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: Protect AI providerId: protectai created: '2026-06-20' modified: '2026-06-20' reconciled: false tags: - AI - ML - Security - LLM - Guardrails - Rate Limiting - Quotas - Throttling description: >- The primary developer surface, the LLM Guard API, is self-hosted: there is no vendor-imposed rate limit. Throughput is bounded only by the resources of the deployment you operate (CPU/GPU, worker count, model load), and operators may add their own gateway-level limits. The commercial platform products (Guardian, Recon, Layer) run as managed services whose limits are governed by account agreements and are not publicly documented. No specific per-endpoint values are reconciled in this artifact. notes: >- Self-hosted LLM Guard has no provider rate limit; tune Uvicorn workers and scanner selection for throughput. Commercial product limits depend on the Protect AI / Palo Alto Networks agreement; verify on reconciliation. sources: - https://llm-guard.com/api/overview/ - https://github.com/protectai/llm-guard - https://protectai.com/ responseCodes: throttled: 429 limits: - name: LLM Guard API (Self-Hosted) scope: deployment metric: requests limit: no provider-imposed limit notes: Bounded by your own compute and worker configuration; add gateway limits as needed. - name: Guardian / Recon / Layer (Commercial) scope: account metric: requests limit: see account agreement notes: Managed-service limits set by contract; not publicly documented. policies: - name: Self-Hosted Throughput description: Scale LLM Guard horizontally and suppress unused scanners to raise effective throughput; no vendor quota applies. - name: Backoff Strategy description: Clients should implement exponential backoff with jitter and honor Retry-After when fronting LLM Guard with a gateway or calling commercial endpoints. maintainers: - FN: Kin Lane email: kin@apievangelist.com