Skip to content

Alternatives

Portkey alternatives in 2026: what to use after the Palo Alto Networks acquisition

Palo Alto Networks completed its acquisition of Portkey on 29 May 2026 and the hosted product is now Prisma AIRS AI Gateway, so stay if runtime AI security is what you are buying. If you chose Portkey as a developer gateway, the alternatives are its own open-source gateway, LiteLLM or Bifrost to self-host, Cloudflare AI Gateway on a platform you already use, Langfuse for observability, and a BYOK control plane such as INFRO for budgets and cost on your own provider accounts.

By INFRO Editorial

Last updated

Last checked

On this page
  1. What changed at Portkey
  2. Stay, or move?
  3. The alternatives, side by side
  4. What each costs at your volume
  5. Which one to choose
  6. How to evaluate the move
  7. What moving to INFRO involves
  8. Questions
  9. Sources

What changed at Portkey

  • Ownership. Palo Alto Networks announced it had closed the acquisition of Portkey on 29 May 2026.
  • The product. Portkey's pricing page now reads "Portkey is now PRISMA AIRS AI Gateway, generally available for all enterprises".
  • The plans. A free Developer plan (10,000 recorded logs a month, 3-day log retention, marked not suitable for production), Production at $49 a month (100,000 logs, 30-day retention, $9 per extra 100,000), and custom Enterprise.
  • The open-source gateway stays MIT licensed. Logging, observability and prompt management are in the hosted app, not the open-source gateway.

An acquisition by a strong company is not a problem by itself. The question is whether the product's direction (now plainly security) is yours, and whether the controls you rely on sit in the tier you buy.

Stay, or move?

If your priority isThen
Runtime AI security: inspecting what models receive and returnStay. That is the product's centre now
You already standardize on Palo Alto NetworksStay. One vendor, one procurement cycle
A self-hosted gateway with no agreementPortkey's open-source gateway, LiteLLM or Bifrost
A free hosted gateway on a platform you useCloudflare AI Gateway
Prompt management and evaluationLangfuse
Budgets and cost per customer on your own provider accountsA BYOK control plane such as INFRO

The alternatives, side by side

Each option is described from its own pricing page and documentation, checked on the date at the top. INFRO is listed first because this is INFRO's site; its limits are listed as plainly as everyone else's.

Who it is for
Teams that hold, or want to hold, their own provider contracts and need routing, failover, budgets and cost per customer across them.
How it charges
A flat plan of $35–$599 a month, chosen by request volume (100K–10M requests a month), never by spend. 0% markup on model spend: your providers bill you directly.

Strengths

  • Every request runs on a provider account you own; INFRO holds none, so nothing falls back onto someone else's contract.
  • Budgets refuse a request before the provider is called, per organization, project, member and key, and (from the Team plan) per end customer.
  • Each response names the provider, connection and model that served it, with an estimated provider cost.

Limits

  • No shared capacity: when every connection you own fails, the request fails with an error that says why.
  • Hosted only on its plans; a self-hosted deployment needs an enterprise agreement.
  • No prompt management or output evaluation.
Who it is for
Teams that liked Portkey's routing and will host the open-source gateway themselves.
How it charges
Production is $49 a month with 100,000 recorded logs, plus $9 per additional 100,000. The open-source gateway is MIT licensed. Enterprise is custom.

Strengths

  • Mature routing, fallbacks, guardrails and observability.
  • An MIT-licensed gateway you can host yourself.

Limits

  • Palo Alto Networks completed its acquisition on 29 May 2026; the product is now Prisma AIRS AI Gateway, so check its roadmap fits a team not buying the wider platform.
  • Logs, observability and prompt management are in the hosted app, not the open-source gateway.

Full comparison with Portkey (Prisma AIRS AI Gateway)

Who it is for
Teams that want no vendor in the request path and have engineers to run a production service.
How it charges
MIT-licensed and free to self-host. Enterprise is a license key on the same image, priced by deployment size; no price is published.

Strengths

  • No third party sees your traffic or your keys.
  • Spend tracked per key, team, user, tag, with a budget and model list on each key.
  • Broad provider coverage and an active open-source project.

Limits

  • You run it: a server, a Postgres database, upgrades, and someone on call while it is in the request path.
  • Its release pipeline is part of your supply chain: two PyPI releases carried credential-stealing code on 24 March 2026.

Full comparison with LiteLLM

Who it is for
Teams that want a self-hosted gateway with budgets and an MCP gateway in one binary.
How it charges
Apache 2.0 and free to self-host.

Strengths

  • Written in Go, started with npx, Docker or as a Go library.
  • Failover between providers, and virtual keys with team and customer budgets.

Limits

  • You run and patch it.
  • Its published overhead figures are its own benchmarks, and two of them disagree.
Who it is for
Applications already on Cloudflare that want logging, caching and spend limits close to their Workers.
How it charges
Core features are free. Your own keys carry no fee; paying through Cloudflare's Unified Billing adds 5% on each purchase.

Strengths

  • Free core gateway with caching, logging, rate limits and model fallbacks.
  • Spend limits that count your own-key traffic for models with known pricing.

Limits

  • Spend limits are eventually consistent: a burst of concurrent requests can briefly pass them.
  • Caching is exact-match only, for text and image responses.

Full comparison with Cloudflare AI Gateway

Who it is for
Teams whose need is tracing, prompt management and evaluation rather than routing.
How it charges
Hobby is free with 50,000 units a month; Core is $29 a month with 100,000 units, then $8 per 100,000. MIT licensed (except enterprise folders) and self-hostable.

Strengths

  • Open-source tracing, prompt versions and evaluation.
  • Owned by ClickHouse since January 2026, and still open source and self-hostable.

Limits

  • Not a gateway: requests do not pass through it, so it cannot route, fail over or refuse a request against a budget.

What each costs at your volume

Portkey's hosted plan is metered by logs, which for a gateway means requests; INFRO's plans are metered by requests too. At the same request volumes, monthly billing:

Requests a monthPortkey ProductionINFRO
100,000$49 ($49 + 0 × $9)$35 (Developer)
500,000$85 ($49 + 4 × $9)$79 (Startup)
2,000,000$220 ($49 + 19 × $9)$199 (Team)
Portkey's figures from its pricing page; INFRO's from its plan ladder. Neither charges a percentage of model spend, and your providers bill model spend either way.

At these volumes INFRO's plan costs less than Portkey's hosted plan. Portkey costs less in two cases: its free Developer plan (10,000 logs a month, not for production), and its MIT-licensed gateway, which costs nothing to license if you run it yourself. Choose on what the plan includes (log retention, budgets that refuse before the call, cost per customer, failover on your own keys), not on price alone. The INFRO vs Portkey comparison works through the difference.

Which one to choose

Stay on Portkey
if AI security is your requirement, or you are standardizing on Palo Alto Networks.
INFRO
if you need budgets that refuse before the call, cost per customer and failover across your own provider accounts, on a flat plan.
Portkey's open-source gateway
if you liked Portkey's routing and will run it yourself, without the hosted logs.
LiteLLM or Bifrost
if you want a self-hosted gateway with budgets and can operate it in the request path.
Cloudflare AI Gateway
if your traffic already runs through Cloudflare and a free core gateway is enough.
Langfuse
if you used Portkey mostly for logs, prompts and evaluation; pair it with a gateway if you need routing.

How to evaluate the move

  1. List what you use: routing configs, virtual keys, budgets, guardrails, prompt templates, logs. Each maps to a different destination.
  2. Check your contract: your renewal date, and what the Enterprise tier would cost for the controls you need.
  3. Export logs and prompts to storage you control before changing anything.
  4. Pilot with real traffic: one service behind a flag for a week; compare errors, latency and cost per request. The Portkey migration guide maps each object.

Questions

Who owns Portkey now?
Palo Alto Networks. It announced the completed acquisition on 29 May 2026, and Portkey's pricing page says the product is now Prisma AIRS AI Gateway.
Is Portkey's gateway still open source?
Yes. The gateway on GitHub is MIT licensed. Logging, observability and prompt management are part of the hosted app, not the open-source gateway.
What is the closest self-hosted alternative to Portkey?
Portkey's own open-source gateway is the lowest-change option. LiteLLM and Bifrost add budgets on virtual keys; all three put a service you operate in the request path.

Sources

Every statement about another company on this page comes from that company's own page, opened on the date shown. Prices and defaults change; check the source before you decide.

  1. 1.portkey.ai/pricing (opens in a new tab) — checked Oct 24, 2026.
    • Portkey's Developer plan is free and records 10,000 logs a month; it is marked not suitable for production.
    • Portkey's Production plan is $49 a month and includes 100,000 recorded logs.
    • Past the included volume, Production adds $9 for each additional 100,000.
    • Portkey keeps logs for 3 days on its free Developer plan and 30 days on Production.
    • Portkey's pricing page says Portkey is now Prisma AIRS AI Gateway.
  2. 2.paloaltonetworks.com/company/press/2026/palo-alto-networks-completes-acquisition-of-portkey-to-secure-ai-agents (opens in a new tab) — checked Oct 24, 2026.
    • Palo Alto Networks completed its acquisition of Portkey on 29 May 2026.
  3. 3.github.com/Portkey-AI/gateway (opens in a new tab) — checked Oct 24, 2026.
    • Portkey's open-source gateway is MIT licensed; logging, observability and prompt management are available on the hosted app only.
  4. 4.litellm.ai (opens in a new tab) — checked Oct 24, 2026.
    • The LiteLLM gateway is MIT licensed and free to self-host.
  5. 5.docs.litellm.ai/docs/enterprise (opens in a new tab) — checked Oct 24, 2026.
    • LiteLLM Enterprise is a license key on the same image, priced by deployment size and usage; no price is published.
  6. 6.docs.litellm.ai/docs/simple_proxy (opens in a new tab) — checked Oct 24, 2026.
    • LiteLLM records the cost of each request and shows spend per key, team, user and tag; each key carries a team, a budget and a model list.
    • The LiteLLM gateway is a self-hosted server started with Docker, Helm or Terraform, backed by a Postgres database you run.
  7. 7.docs.litellm.ai/blog/security-update-march-2026 (opens in a new tab) — checked Oct 24, 2026.
    • On 24 March 2026, litellm 1.82.7 and 1.82.8 on PyPI carried credential-stealing code for about 40 minutes; LiteLLM says the official Proxy Docker image was not affected and 1.83.0 is clean.
  8. 8.github.com/maximhq/bifrost (opens in a new tab) — checked Oct 24, 2026.
    • Bifrost is an Apache 2.0 open-source gateway written in Go, started with npx, Docker or its Go SDK.
    • Bifrost lists failover between providers and models, and virtual keys with team and customer budgets, as well as an MCP gateway.
  9. 9.developers.cloudflare.com/ai-gateway/reference/pricing (opens in a new tab) — checked Oct 24, 2026.
    • Cloudflare AI Gateway's core features are free.
  10. 10.developers.cloudflare.com/ai-gateway/configuration/bring-your-own-keys (opens in a new tab) — checked Oct 24, 2026.
    • Cloudflare AI Gateway can store your own provider keys (BYOK) in the dashboard.
  11. 11.developers.cloudflare.com/ai-gateway/features/unified-billing (opens in a new tab) — checked Oct 24, 2026.
    • Usage paid for through Cloudflare's Unified Billing carries a 5% fee on each purchase.
  12. 12.developers.cloudflare.com/ai-gateway/features/spend-limits (opens in a new tab) — checked Oct 24, 2026.
    • Cloudflare's spend limits count both Unified Billing and BYOK requests, for models with known pricing, and answer 429 when exceeded.
    • Cloudflare's spend limits are eventually consistent: a burst of concurrent requests can briefly pass the limit.
  13. 13.developers.cloudflare.com/ai-gateway/configuration/fallbacks (opens in a new tab) — checked Oct 24, 2026.
    • Cloudflare AI Gateway can fall back to another model or provider through its Universal endpoint when a request returns an error.
  14. 14.developers.cloudflare.com/ai-gateway/features/caching (opens in a new tab) — checked Oct 24, 2026.
    • Cloudflare caches only identical requests, for text and image responses, for 60 seconds to one month; semantic caching is planned, not available.
  15. 15.langfuse.com/pricing (opens in a new tab) — checked Oct 24, 2026.
    • Langfuse Cloud's Hobby plan is free with 50,000 units a month and 30 days of data access.
    • Langfuse Core is $29 a month with 100,000 units included; overage is $8 per 100,000 units, lower with volume.
  16. 16.github.com/langfuse/langfuse (opens in a new tab) — checked Oct 24, 2026.
    • Langfuse is MIT licensed except for its enterprise folders, and can be self-hosted with Docker.
  17. 17.langfuse.com/blog/joining-clickhouse (opens in a new tab) — checked Oct 24, 2026.
    • ClickHouse acquired Langfuse in January 2026; Langfuse says it stays open source and self-hostable.
  18. 18.langfuse.com/resources/engineering/llm-gateway (opens in a new tab) — checked Oct 24, 2026.
    • Langfuse is an observability and evaluation platform, not a gateway: requests do not pass through it on the way to a provider.

The team building INFRO, a BYOK-first control plane for production AI. Articles are written from the engineering work on the gateway and cite their external sources.