AI gateway in Palo Alto Networks' Prisma AIRS security platform
INFRO vs Prisma AIRS AI Gateway (Portkey)
Portkey now ships as Prisma AIRS AI Gateway: Palo Alto Networks completed its acquisition in May 2026 and has made the gateway part of its Prisma AIRS AI security platform. It is a mature, widely deployed AI gateway with a very large model catalog, an MIT-licensed open-source gateway you can still self-host, prompt management, and a security surface — guardrails and PII redaction — that is among the strongest in this category. If the requirement driving your evaluation is securing and governing what AI models are allowed to receive and return, it is the serious answer and you should look at it properly.
- Runtime AI security is the requirement. Guardrails, PII redaction and content controls on the request and the response are their centre of gravity, and now sit inside a security company's platform. INFRO's guardrails are pattern-based checks for PII, secrets and your own terms on chat completions, plus classifier checks for prompt injection, jailbreaks and harmful or off-topic requests that read prompts only, run on your own connection to a decision model's provider, and come with the Business plan. There is no image or audio inspection.
- You are already standardising on Palo Alto Networks. Buying the gateway from the vendor that holds the rest of your security estate is a coherent decision and one procurement cycle rather than two.
- You want an open-source gateway you can read and run yourself, with no enterprise agreement. Their gateway is open source; INFRO runs in your own infrastructure only under an enterprise agreement.
- Prompt management and versioning matter to you as a product surface. That is theirs, not ours.
- You need the widest possible model catalog, including long-tail and community models we do not carry.
This section is not a courtesy. If one of these describes you, pick Prisma AIRS AI Gateway (Portkey) — a customer who adopts the wrong layer on our advice is worse for us than one we never had.
Where the two actually diverge.
- Security-first and cost-first are different products
- Their organising question is what an AI request is allowed to contain. Ours is what it costs, whether it succeeded, and who in the company owns that. Both products route, and both observe — but the surface each one invests in is the one its buyer is measured on, and a security team and a platform-and-finance team are not measured on the same thing.
- AI FinOps is the whole product here, not a dashboard tab
- Every request carries an estimated provider cost — from the vendor's published rate, or from the rates you negotiated once the Team plan lets you enter them — attributed to a project, key, model, route or end user. Budgets act on that number at four levels. On the Business plan the month's estimate is checked against your provider's own cost API and the variance is stated rather than assumed. That chain — estimate, attribute, govern, reconcile — is what INFRO is for.
- A subscription, not a meter
- INFRO's plan price is fixed and its request allowance is an allowance, not a per-request charge: exceeding it is counted and shown in the console, which offers the larger plan, and never drops production traffic. Nothing INFRO sells is priced per token, and INFRO takes 0% of model spend. Read any gateway's current pricing page before you decide — including ours — because the structure is the part that decides what a busy month costs you.
- Spend governance on the entry plan, not behind an agreement
- Budgets that refuse a request at organization, project, member and key level, model allowlists at organization, project and key level, and the audit log are on every INFRO plan including the entry one; members and roles come with the Startup plan, since Developer is a single seat. The ladder above that is published: negotiated rates and end-customer caps from Team, and single sign-on, SCIM, guardrails, the MCP gateway and reconciliation on Business. Prisma AIRS AI Gateway's pricing page lists granular budgets and rate limits, and single sign-on, on Enterprise. The judgement behind INFRO's line is that a company needs to cap and audit spend on day one, at the scale it has on day one — not after it has grown enough to sign an enterprise agreement.
Side by side.
Our column carries dashes too. A comparison where one side is all checks is one nobody believes, and rightly.
| Capability | Prisma AIRS AI Gateway (Portkey) | INFRO |
|---|---|---|
| One API across many providers | yesvery large catalog | yes |
| Routing, retries and failover | yes | yes |
| Your own provider accounts | yes | yesconnected per environment |
| Request logging and traces | yes | yes |
| Guardrails and PII redaction | yestheir specialism | partialchat patterns plus prompt classifiers; Business plan |
| Prompt management | yes | nonot built |
| Cost attributed by project, key and end user | yesvia request metadata | yes |
| Estimates at your negotiated rates (Team and above) | partial | yesper-org rate overrides |
| Reconciliation against the provider's own bill | no | partialOpenAI and Anthropic, Business plan; variance stated |
| Budgets, allowlists and audit log on the entry plan | partialgranular budgets on Enterprise | yes |
| Image, video, and audio | partialimage and audio; no video documented | yes |
| Self-hosting | yesopen-source gateway, MIT | partialenterprise agreement |
| Available today | yes | noprivate early access |
Our reading of each product's public documentation as of October 2026. We do not quote competitors' prices — theirs change and a wrong number about someone else's business is worse than none. Think a row is wrong? Tell us and we will fix it.
Work out what the difference is worth
Estimate what your current model mix costs at your providers' list rates and the plan it needs, or have us read your real usage and tell you where we would and would not help.