AI Gateway HQ vs. Portkey / Prisma AIRS AI Gateway
Use a managed gateway that blocks unfunded requests and keeps prompt text out of the request ledger by default.
- Pre-dispatch control
- Operator-ready evidence
- Prompt storage off by default
Choose AI Gateway HQ when the gateway must prevent an unreserved request from reaching a provider and give finance, security, and operations one minimal-data evidence trail. Portkey remains a credible alternative for teams prioritizing its broader MCP, caching, and observability surface.
Every strict request reserves organization and workload exposure before a provider call; the ledger then records the route, reason, upstream attempts, cost, and settlement without storing prompt bodies by default.
Compare what each gateway does.
Our column describes the working product. The Portkey / Prisma AIRS AI Gateway column summarizes its linked documentation and pricing; it does not rely on anonymous review scores.
A compatible endpoint, encrypted BYOK pools, health-aware routing, shared cooldowns, and reason-coded fallback.
Universal API with broad providers, caching, MCP, retries, fallbacks, circuit breaking, load balancing, and canary routing.
Budget and capacity are reserved before forwarding. Settlement uses supported provider-reported usage, and promotional credit cannot fund server-paid models.
Budgets, rate limits, and usage/cost controls; the reviewed docs do not promise the same atomic pre-forward wallet reservation.
OIDC, required MFA, least-privilege roles, workload keys, signed context, staged enforcement, and local attack screening.
Enterprise SSO, RBAC, guardrails, and substantially broader current observability and administration.
AWS serverless deployment behind WAF, tenant-bound encryption, signed releases, metadata-only request records by default, tamper-evident exports, and approved support access.
Managed and open-source/self-hosted options; Enterprise lists private/VPC deployment and data export.
A free BYOK proving tier, then $0.10 per 1,000 successful Flex requests. Inference bought through customer-owned accounts has no percentage markup.
Free includes 10,000 logs and three-day retention; Production lists $49/month for 100,000 logs plus usage overage; Enterprise is custom.
Consider Portkey when an already-shipping MCP gateway or semantic cache is more important than deterministic prepaid enforcement.
We reviewed Portkey / Prisma AIRS AI Gateway's linked documentation and pricing on August 7, 2026. Vendors measure requests, logs, credits, infrastructure, and capacity differently, so compare the underlying units and recheck current terms before purchasing.