Portkey alternatives

The right Portkey alternative depends on the job: LiteLLM for a self-hosted gateway with virtual keys, OpenRouter for one bill across many models, Helicone for request logs, and RelayPlane for stopping a runaway agent run. Cloudflare AI Gateway fits teams that already run on Cloudflare.

We make RelayPlane, so we say where the other options fit better.

How we compared them

We wrote down four questions before looking at any product. Each tool below is judged on these and nothing else.

  • Where it runs: on your machine, in your own cloud account, or on the vendor’s servers.
  • What the tool costs, apart from the model spend you already pay.
  • Spend control: whether it can stop a run, or only show you the bill afterwards.
  • Lock-in: whether you can read the source and move away.

Facts about other companies come from their own pages, checked on 2026-10-19. Anything marked “in our view” is our opinion. The options are listed in alphabetical order, with no ranking.

Where Portkey fits

Portkey is the tool you are leaving, so here is what it does well. Its open source gateway says “Route to 250+ LLMs with 1 fast & friendly API” and runs locally with npx @portkey-ai/gateway. Its hosted plans add logs and prompt management.

The cost is in the plan ladder. The Developer plan allows 10k recorded logs per month and Portkey describes it as “Not suitable for production workloads.” The Production plan is $49/month. Budget and rate limits appear in the Enterprise governance row of its pricing page, so a team that wants hard spend limits may end up on a custom plan.

Cloudflare AI Gateway

Cloudflare describes AI Gateway as “Available on all plans” and lists analytics and logging, plus caching and rate limiting. Its analytics show “the number of requests, tokens, and the cost it takes to run your application.”

It runs on the vendor's network and its docs describe rate limiting as limiting the number of requests your application receives. In our view it fits a team already on Cloudflare that wants visibility and caching. Its overview page does not mention a dollar cap on one agent run.

Helicone

Helicone is an open source LLM observability platform. Its Hobby plan includes 10,000 free requests and the Pro plan is $79 per month. It is built to log and analyze requests, which suits a team debugging prompts and quality.

Logging after the fact is a different job from refusing a request before it goes out. Our Helicone comparison goes through that split.

LiteLLM

LiteLLM's docs call it “a single, unified interface to call 100+ LLMs” and list a “Self-hosted LLM Gateway (Proxy) with virtual keys, cost tracking, and an admin UI.” It also has “Built-in retry / fallback logic across multiple deployments via the Router.”

This is the closest swap for Portkey if you want to host the gateway yourself. The price is operating it, since you run the proxy and its admin UI. Teams with many users and keys get the most from virtual keys.

OpenRouter

OpenRouter is a hosted service with 500+ models on its Standard plan. Its FAQ says “Inference is billed at the provider's list price on every plan,” and its fee is a platform fee of 5.5% on Standard. Budgets and spend controls are marked Yes on Standard and No on Free.

It fits a team that wants one invoice and the widest model list with no gateway to run. The tradeoff is that your traffic goes through their servers and the fee scales with what you buy.

RelayPlane

RelayPlane is a free, open source (MIT) proxy that runs on your own machine. It is built around what each agent run costs. A cap is checked before a request is forwarded, and the request that would cross the cap is the one that gets blocked, with a budget_exceeded response. The budget cap docs show the full config, and the onBreach setting picks what happens next.

npm install -g @relayplane/proxy
relayplane start
relayplane cap set --day 10

The limits are plain. It is not a hosted service, so there is no vendor console, and a team that needs enterprise controls or the widest provider list is better served by one of the options above. See the LLM proxy page and the RelayPlane vs Portkey comparison for the detail.

Which one for whom

To host a gateway for a team of users, start with LiteLLM. For no infrastructure and one bill, OpenRouter is the simpler path. Helicone answers what your prompts did, and a team already on Cloudflare can try its gateway first. If the question is what a single agent run may spend and how to stop it, in our view a local proxy with a cap is the shorter path, and many teams run it next to a logging tool.

Try the quickstart

Sources