Updated
AI models and media

LLM Gateway & Routing

Put one layer between your app and the model providers, so you can switch models, fail over and track spend without rewriting calls.

Ask your AI about this, with this page as the source:ChatGPT ↗Claude ↗Perplexity ↗

Trends

OpenRouter gained the most ground in package downloads: 5.1× against the median npm package over a year.

Public counts each month: pull requests on GitHub, package downloads, stars, Product Hunt followers. They show direction, not market share.

ToolDownloadsStarsProduct Hunt
Vercel AI Gateway23.3M/wk——
Cloudflare21.3M/wk3.1× vs npm—7.8k followers
LiteLLM20.9M/wk★ 61k215 followers
OpenRouter2.8M/wk5.1× vs npm—1.1k followers
Portkey184.6k/wk+25% vs npm★ 13.2k—
RespanRenamed665/wk★ 1—
Kong Gateway—★ 44.3k—
UnifyArchived———

The real choice

Hosted gateway or self-hosted proxy. A hosted gateway gives you one key and one bill in minutes, but every request passes through someone else's service. A self-hosted proxy keeps your provider keys and traffic on your own servers, and you run and update it.

Pick by situation

Tap the ones that are you — the tools that fit light up below.

Compare the tools

Grouped by the side of the choice they answer, not ranked. Open a row for when to use it, the trade-off and what makers say.

ToolBest for
Hosted model marketplace2
OpenRouterapi · free tierTrying many models from many providers with one API key and one bill, without opening an account at each.No evidence yet2.8M/wk5.1× vs npm

Use it whenYou want to compare models quickly or offer users a model picker.

Trade-offA fee on top of provider prices, and your traffic and data pass through a third party.

llms.txtIn Replit

No maker's product we track shows it for this yet.

Vercel AI Gatewayservice · free tierOne key and one bill for hundreds of models at provider list prices, with fallbacks and budgets, built into the AI SDK.No evidence yet23.3M/wk

Use it whenYou're on Vercel or the AI SDK and want provider switching and failover without another vendor account.

Trade-offTied to a Vercel team and prepaid credits, and bringing your own provider keys needs the paid tier.

llms.txtMCPCLIIn v0

No maker's product we track shows it for this yet.

Self-hosted proxy1
LiteLLMlibrary · open source · free tierRunning your own OpenAI-compatible proxy in front of your own provider keys, with retries, routing and per-key budgets.No evidence yet20.9M/wk

Use it whenYou want keys and request data to stay on your infrastructure.

Trade-offYou deploy, monitor and upgrade it yourself; SSO and audit logs are in the paid enterprise tier.

llms.txtMCP

No maker's product we track shows it for this yet.

Managed gateway + controls2
Heliconeplatform · open source · free tierAdding a gateway when your main goal is logging and debugging every request, with caching and rate limits on the side.13.8k/wk

Use it whenSeeing every request, its cost and latency matters more to you than advanced routing.

Trade-offIts focus is observability; routing features are thinner than a dedicated gateway's.

llms.txt

Used by Pretty Prompt.

Helps us debug AI, and have clarity on AI analytics, costs, and latency. We've also been able to save quite a lot of credits because of its caching functionality!
Pretty Prompt, the makerSep 2026 ↗
Portkeyplatform · open source · free tierA gateway that also handles guardrails, prompt management and request logs in one dashboard.No evidence yet184.6k/wk+25% vs npm

Use it whenYou want routing and governance rules without building them, with the option to self-host the open-source gateway.

Trade-offThe free tier caps logged requests; the full control layer is a paid hosted product.

llms.txtCLI

No maker's product we track shows it for this yet.

Before you choose

How to approach it

Write your code against the OpenAI-compatible API shape from the start; every option here speaks it, so switching later is a base-URL change. Set a fallback model and a spend limit before launch, not after the first outage or surprise bill. If you only call one provider, you can skip this layer until you add a second.

Common mistakes
  • Adding a gateway with no fallback model configured, so it becomes one more thing that can go down rather than the thing that keeps you up.
  • Sending user data through a hosted gateway without checking its retention settings and adding it to your list of subprocessors.
  • Treating a gateway's per-key budget as optional and only discovering a runaway agent loop on the monthly invoice.

Other options

Real choices most makers here won't need to weigh.