← Blog

Portkey Alternatives: 9 AI Gateways After Palo Alto

November 16, 2026

AI CodingTool Comparisons

The best Portkey alternative depends on which part of Portkey you actually use. If you want an open-source gateway you host yourself, pick LiteLLM or Bifrost. If your platform team runs Kubernetes and Envoy, pick Agent Router. If you need governance, guardrails and private deployment for a large company, look at TrueFoundry or Kong AI Gateway. If you'd rather not run anything, use Cloudflare AI Gateway or Vercel AI Gateway. If you want one key and one bill for hundreds of models, use OpenRouter.

Portkey sells five things in one product: an AI gateway, observability, guardrails, prompt management and governance. Most alternatives cover two or three of those. So the first step is deciding which ones you'd miss.

The second reason people are searching: Portkey changed owners this year. Here's what happened, then a comparison of nine options.

What Changed at Portkey in 2026

The facts, from primary sources:

The open-source side is quieter. The Portkey gateway repo is MIT-licensed and not archived. But when we checked on November 16, 2026, its main branch had no commits after May 25, 2026, and the newest tagged release was v1.15.2 from January 2026. The README still announces a Gateway 2.0 pre-release that merges Portkey's enterprise gateway into open source; that branch's latest commit was in March 2026.

None of this means Portkey is shutting down. It means the product now lives inside a large security vendor, with an enterprise focus. If you chose Portkey as a startup-friendly gateway or for its open-source core, it's fair to check your options.

Quick Comparison

OptionShapeLicenceWho runs itClosest Portkey feature
Portkey (baseline)Gateway, observability, guardrails, promptsMIT gateway; hosted productYou, or Portkey's cloudAll five pieces
LiteLLMPython SDK plus proxyMIT outside enterprise/YouGateway, virtual keys, budgets
BifrostStandalone gateway with web UIApache-2.0YouGateway, fallbacks, caching
Agent RouterKubernetes control plane on EnvoyApache-2.0You, or TetrateGateway, MCP routing
TrueFoundryEnterprise AI, MCP and agent gatewayCommercialCloud, VPC, on-premGovernance, guardrails, prompts
Kong AI GatewayAI plugins on KongCommercialKonnect or on-premGuardrails, semantic cache, budgets
Cloudflare AI GatewayManaged gatewayHosted serviceCloudflareLogs, caching, rate limits
Vercel AI GatewayManaged gatewayHosted serviceVercelRouting, failover, budgets
OpenRouterHosted model marketplaceHosted serviceOpenRouterUniversal API
HeliconeGateway plus loggingApache-2.0You, or HeliconeObservability

Open-Source Gateways You Host Yourself

Pick from this group if you used Portkey's open-source gateway, or if your security team wants traffic and keys inside your own network.

LiteLLM: the widest provider coverage

Best for: teams that want one OpenAI-format endpoint for every provider, with spend controls.

LiteLLM calls itself an open-source AI gateway for 100+ LLMs. You can use it as a Python SDK inside your code, as a central proxy server, or both. The README lists virtual keys, spend tracking, guardrails, load balancing and an admin dashboard in the open-source gateway, plus MCP tool support.

Where it beats Portkey: provider breadth, a large community, and an open-source project that's clearly under active development.

Where it's weaker: LiteLLM's enterprise page puts SSO and SCIM, audit logs, secret managers and IP allowlists behind a licence key. Prompt management is thinner than Portkey's. Two PyPI releases were compromised in March 2026, so pin versions. Our LiteLLM alternatives guide covers that incident in detail.

Bifrost: a fast gateway that runs as one service

Best for: teams that want LiteLLM-style proxy features without a Python dependency.

Bifrost is an Apache-2.0 gateway from Maxim. Its README says it puts 23+ providers behind one OpenAI-compatible API. You start it with npx or Docker and configure it in a web UI, with automatic failover, load balancing, semantic caching and budgets through virtual keys.

Where it beats Portkey: simple to stand up, and Maxim's site now leads with Bifrost, so the gateway is the main product. It publishes latency benchmarks, but they're vendor numbers, so test on your own traffic.

Where it's weaker: the README lists guardrails, the MCP gateway, adaptive load balancing and clustering under enterprise deployments. If you chose Portkey for guardrails, that's the line to check.

Agent Router: Envoy for AI traffic

Best for: platform teams that already run Envoy and Kubernetes.

Agent Router is the project formerly called Envoy AI Gateway, now part of the Agentic AI Foundation. The site says it's the same code and the same maintainers. It gives agents one OpenAI-compatible API for models and one router for MCP tools, while platform teams set providers, credentials, limits and failover. Tetrate offers a hosted version if you don't want to run it.

Where it beats Portkey: vendor-neutral foundation governance and an Apache-2.0 licence.

Where it's weaker: it's infrastructure. There's no prompt library or evaluation layer, and without a platform team it's a lot of machinery.

Enterprise Governance Platforms

These are the closest match if you used Portkey for org-wide control: who can call which model, at what budget, with which guardrails.

TrueFoundry: gateway plus MCP and agent governance

Best for: regulated companies that need the gateway in their own VPC, on-prem or air-gapped.

TrueFoundry's AI Gateway sits alongside an MCP gateway, an agent gateway and a prompt registry. Its docs list exact and semantic caching, cost tracking, access control policies, guardrail policies with external providers, and request logging. The site has a free starting tier and enterprise deployments.

Where it beats Portkey: VPC, on-prem and air-gapped deployment sit front and center on its site, and it covers model, MCP and agent traffic in one place.

Where it's weaker: it's a broad platform with deployment and training tools attached. Smaller teams may only want the gateway.

Kong AI Gateway: if you already run Kong

Best for: organisations whose APIs already go through Kong.

Kong AI Gateway adds AI plugins to the gateway you may already operate. The docs list a semantic cache, a prompt compressor, prompt guards (including natural-language jailbreak detection), a sanitizer that redacts PII before it reaches the provider, and AI cost management with budgets. You can run it through Kong's Konnect service or on-prem, and it also handles MCP and agent-to-agent traffic.

Where it beats Portkey: one control plane for AI and non-AI APIs, and mature tooling for platform teams.

Where it's weaker: AI Gateway 2.x moves to a new entity model, so older setups need a migration. It's also more product than a small team needs.

Managed Gateways You Don't Operate

Cloudflare AI Gateway

Best for: adding analytics, caching and rate limits in front of provider accounts you already have.

Cloudflare AI Gateway works on all Cloudflare plans and takes one line of code to start. You get analytics, logging, caching, rate limiting, retries and model fallback. Cloudflare's pricing notes say the core features are free; logs beyond your plan's limits and some extras are billed. Its docs also list data loss prevention and guardrails.

Where it beats Portkey: nothing to host, and it runs on a network you may already use.

Where it's weaker: your prompts pass through Cloudflare, and prompt management isn't part of it.

Vercel AI Gateway

Best for: app teams that want routing, failover and budgets with no infrastructure.

Vercel AI Gateway is a hosted API you can call from any cloud or language. Your app doesn't need to run on Vercel. The docs say it adds zero markup to provider token prices, including when you bring your own keys, and it supports ordered provider and model fallbacks.

Where it beats Portkey: very little setup, and no gateway to operate.

Where it's weaker: budgets only cover spend billed through Vercel's keys, and caching is provider prompt caching. Full-response caching is left to your app.

OpenRouter: one key for many models

Best for: teams that want hundreds of models on one account, without managing provider contracts.

OpenRouter is a hosted model marketplace. Its homepage lists 600+ models from 80+ providers behind one API, with routing and fallbacks across providers. That's a different job from a gateway in front of your own keys, though it supports bring-your-own-key too. Our OpenRouter alternatives guide compares it with similar services.

Where it beats Portkey: the fastest way to try many models, with one bill.

Where it's weaker: less governance, no self-hosting, and traffic goes through OpenRouter's accounts.

A note on Helicone

Helicone shows up on most Portkey comparison lists because it pairs an Apache-2.0 gateway with request logging. But its ownership has moved twice this year. Mintlify announced it was acquiring Helicone in March 2026, and as of today Helicone's site says it is now part of Experiential Labs, with "details to come soon." Its homepage still invites sign-ups. We cover the full timeline in Helicone alternatives. Wait for those details before you move a production gateway there.

How to Choose a Portkey Alternative

What you used Portkey forLook atWhy
Open-source gateway in your networkLiteLLM, BifrostActive projects, self-hosted
Kubernetes-native routingAgent RouterEnvoy-based, foundation-governed
Org-wide governance and guardrailsTrueFoundry, KongRBAC, budgets, PII redaction, private deploys
Logs, caching and rate limits onlyCloudflare AI GatewayFree core features, nothing to host
Failover and budgets for an appVercel AI GatewayHosted, no markup on tokens
Many models on one billOpenRouterMarketplace, not a gateway

Before you switch, list every Portkey feature your code touches: configs, virtual keys, guardrail hooks, prompt templates, metadata on logs. Gateway routing is easy to replace, since most options speak the OpenAI format. Prompt templates and guardrail hooks take the most rework.

Then run the same traffic through two options for a week. Compare added latency, error handling during a provider outage, and whether the logs answer your usual debugging questions.

When Portkey Is Still the Right Call

Stay on Portkey, or move to Prisma AIRS AI Gateway, if:

  • Your company already buys from Palo Alto Networks. A gateway inside your security vendor's platform is easier to approve.
  • You need all five pieces in one product. Few alternatives match gateway, observability, guardrails, prompt management and governance together.
  • Security is the main reason you have a gateway. That's where Palo Alto is taking the product.

Where Taku Fits

Gateways matter when you're shipping your own AI product and need control over every model call. A lot of people just want to use AI apps that someone else has already built and wired up. Taku isn't an AI gateway and doesn't replace Portkey or anything on this list. That's where Taku comes in.

Taku Apps and Sites, built by the best AI minds. Find one, run it, and keep it in your Mac Dock. Taku is in Beta, and the Mac app is available now.

FAQ

Who owns Portkey now?

Palo Alto Networks. It closed the acquisition on May 29, 2026, and Portkey's site now says Portkey is Prisma AIRS AI Gateway.

Is the Portkey gateway still open source?

Yes. The repository is MIT-licensed and not archived. When we checked on November 16, 2026, its main branch hadn't had a commit since May 25, 2026, so watch where development happens before you depend on it.

LiteLLM vs Portkey: which is better?

LiteLLM is the better pick for a self-hosted, open-source gateway with wide provider coverage and an active project. Portkey is stronger if you want observability, guardrails and prompt management in one hosted product, and you're comfortable with Palo Alto's enterprise direction.

What's the best open-source LLM gateway?

LiteLLM for provider coverage, Bifrost for a simple standalone service, and Agent Router for Kubernetes and Envoy shops. All three are open source and self-hostable.

Is there a Portkey alternative with built-in observability?

TrueFoundry and Kong both pair the gateway with logs, metrics and cost tracking. Cloudflare AI Gateway gives you analytics and logs with nothing to host. Helicone also combines a gateway with logging, but check its ownership status first.