Why you're leaving Tyk decides where to go: pick a fully hosted gateway if running one has become a chore, an open-source or Kubernetes-native gateway if you want to self-run without a licence, or a management platform if you need governance across many teams. Settle that one fork and ten options shrink to two or three.
Teams rarely walk away from a gateway because it's slow. They walk away because the bill stopped matching the way they use it, or because keeping the thing alive turned into somebody's second job. Tyk prices its Core plan by the number of APIs you manage and its Professional plan by data planes and environments, so a company that ships lots of small APIs, or spins up a fresh environment per customer, feels that model in a very particular place. Write down which pain brought you here before you read a single name below. It'll save you a week of demos.
Where the field splits
Hosted or self-run. A fully hosted gateway means no clusters to patch, no control plane to upgrade and no on-call rota for the proxy layer. You pay for that with less control over where traffic flows and a bill that grows with usage. Self-running an open-source gateway flips it: no licence, but your team owns the uptime, the upgrades and the 2am alerts. Neither is cheaper in every case, and anyone who tells you otherwise hasn't counted engineer hours.
Kubernetes-native or standalone. If every service you run already lives in Kubernetes and you manage config through Git, a gateway that speaks the Kubernetes Gateway API will feel like part of the furniture. If you've got a mix of VMs, serverless functions and a couple of clusters, a standalone gateway with its own config is often simpler to reason about.
A fast gateway or a whole suite. Some teams need routing, auth, rate limits and nothing else. Others need a developer portal, analytics, API keys for outside customers and governance across several business units. Buying a suite when you need a proxy wastes money; buying a proxy when you need a suite means bolting on four other tools later.
Hosted gateways and management platforms
Kong Gateway
An API and AI gateway with a managed control plane, aimed at teams pushing LLM traffic who want token-based rate limits and semantic caching.
Zuplo
Small, hosted and programmable, it handles API, AI and MCP traffic and hands you a developer portal without a separate project to run.
Gravitee
When the headache is governance across several gateways, teams and consumers, this management platform also takes in event streams and AI agent management.
API7 Enterprise
Built by the people who created Apache APISIX, an enterprise gateway you can run yourself or have hosted, with a separate AI gateway called AISIX.
Amazon API Gateway
Already on AWS? Its fully managed service builds HTTP, REST and WebSocket APIs and charges per request rather than per gateway.
Gateways you run on your own infrastructure
KrakenD
Performance-minded teams self-run this open-source gateway for free, and a paid Enterprise Edition adds SLAs and AI routing on top.
Traefik Hub
Kubernetes shops that manage everything through Git will recognise the model: gateway and API management built on Traefik Proxy, driven by GitOps.
Kgateway (Solo.io)
An Envoy-based gateway for Kubernetes that you operate yourself, with Solo.io standing behind it as the enterprise vendor.
Envoy Gateway
A community project implementing the Kubernetes Gateway API, so you can put Envoy Proxy to work as your gateway with no vendor in the loop.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
DreamFactory
Not a pure proxy at all: it generates governed REST APIs straight from your databases, complete with role-based access, API keys and live OpenAPI docs.
What a gateway costs to run
Budget anywhere from about $25 a month for the smallest hosted plans to the low thousands of dollars a month for the enterprise tiers. That spread comes from the six vendors here that print a price on their own pricing pages this month, and they meter in at least four ways: a flat monthly plan, a plan plus usage, a fee per gateway group plus a per-call charge, and pure pay-per-request, where Amazon's HTTP APIs drop under a dollar per million calls at the highest volume tier (its REST APIs cost more per call).
The other four publish no figure. Where there's an open-source edition there's no licence fee, and the price of any paid or enterprise edition isn't shown on the pages read for this list, so expect a sales call.
The open-source gateways don't charge a licence, but they aren't free. Budget for the cluster they run on, the time to upgrade them, and the person who gets paged. If you're comparing a hosted plan against self-running, put those hours into the sum before you call the open-source route cheaper. Also check whether a vendor's bill is driven by APIs, environments or calls, because that's the number that moves when your business grows. If you also sell API access to outside developers, the RapidAPI alternatives list covers the distribution side.
Do you need an AI gateway as well?
This is the change that reshuffled the category. LLM routing, token budgets and MCP support are being added by most gateway vendors, Tyk included, along with Kong, Zuplo, KrakenD, Traefik and Gravitee. So the question has stopped being "does it do AI" and become "does it do the AI parts I'll use".
If your product calls one model provider from one service, you probably don't need much: a rate limit and a secret store will do. Once several teams call several models, you'll want per-team token budgets, routing between providers, and a log of which prompt cost what. If agents inside your company call internal tools over MCP, the gateway becomes the place you decide which agent may touch which API. Ask each vendor to show that last part working, not on a slide. For one vendor's take, see this review of Kong's AI gateway and its rivals.
Do you make API gateways or API management platforms? Tool makers can ask to be considered for a featured place on this list.
Mistakes that make a gateway move hurt
- Migrating every route at once. Move one low-risk API, run both gateways side by side, then move the rest in batches.
- Forgetting the plugins. Custom middleware you wrote for Tyk won't port. List every one before you sign anything and check each has an equivalent.
- Picking on the AI features alone. They're changing every quarter. Choose on the routing, auth and ops model you'll live with for years.
- Ignoring who edits config. A GitOps gateway is great for platform engineers and painful for a product team that liked clicking in a dashboard.
- Skipping the exit test. If config can't be exported cleanly, you're buying the same lock-in you're trying to escape.
Where I'd land for each kind of team
Small team, no platform engineers, want it gone from your worry list: a hosted programmable gateway. Everything already in Kubernetes and managed through Git: one of the Envoy-based or Traefik-based gateways, and stay with the open-source edition until you need support. Lots of business units and outside API consumers: a full management platform with governance and a portal. All-in on one cloud: that cloud's own managed gateway, and accept the per-request bill. If what you really want is APIs on top of existing databases, a generator beats a proxy.
Before you sign: a one-hour bake-off
Don't judge a gateway from its docs, and don't run this on five candidates. Pick your two finalists and run the same test on both:
- Take one real API you already expose through Tyk, ideally one with auth and a rate limit, and recreate it on each candidate.
- Add a key-based auth rule and a limit of a handful of requests per second, then hammer it with a simple load tool and check the excess calls get rejected with a clear error.
- Change one route through whatever the vendor calls config (a dashboard, a YAML file in Git, a Kubernetes resource) and time how long the change takes to go live.
- Break something on purpose: point the upstream at a dead host and see what the client receives and what the logs tell you.
- Export the config. If you can't get your setup back out in a readable format, you've found your next migration problem early.
The scores that matter here are how long step three took and how readable step four's logs were. Raw throughput rarely decides it at the volumes most teams run. One afternoon of breaking things will teach you more than any comparison grid.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.