any harness, any tool

Keep your tools and your harness. Cut what they cost.

Your engineers already have the tools they like: Claude Code, Codex, Cursor, Cline, the OpenAI and Anthropic SDKs, and the agents and pipelines you built yourselves. Any harness can be integrated inside the proxy. For most tools it is one base URL change. Nobody re-learns anything, and every request goes through a proxy tuned to cost as little as it can, with the same quality.

Works with any harness your team already uses

One base URL. Same tools. A smaller bill.

Claude Code
Codex, Cursor, Cline
Your own harness

features

Claude Code for teams, through your company's LLM gateway

Claude Code officially supports routing through an organization's LLM gateway. In Anthropic's own words, "the provider key stays server-side; developers hold gateway credentials instead." With Proxen that gateway credential is the engineer's own key. Set the base URL and the key once in the environment, and Claude Code runs exactly as before, on your company's Anthropic account. Underneath, the proxy trims long histories, picks the right model per task and keeps the cache warm, which is where most Claude Code cost goes.

Terminal showing Claude Code started with a gateway base URL and a personal key set in the environment.Illustrative example

features

Codex, Cursor, Cline and the other editors

Every tool that lets you set a custom endpoint works the same way. Codex, Cursor, Cline, Roo Code, Continue, Zed, Aider and OpenCode each have a base URL field and a key field. Put the Proxen address in one and the engineer's key in the other. Model names stay the same in the tool. The routing rules we set with you decide which model answers each task, and the log shows which one did. Mixed teams on Cursor and Codex land in the same ledger as Claude Code.

editor settings panel with a custom base URL and API key field filled in for Proxen.Illustrative example

features

SDKs, your own agents and pipelines

The OpenAI and Anthropic SDKs for Python, Node and Go take a base URL in their constructor or from an environment variable. So do LangChain, LlamaIndex, the Vercel AI SDK, n8n and a notebook with curl in it. Change that one value and your own harness runs through the proxy, under its own key with a cap. If your agent framework needs something more, we integrate it inside the proxy with you. Unattended pipelines are often the best place to save: much of their work can wait for batch at half price.

Code sample creating an SDK client with a Proxen base URL and a per-agent key.Illustrative example

Why Proxen

The proxy wraps around how you already work

Proxen sits between the tools your engineers already run and the providers your company already pays. It speaks the same API the tools expect, so the rollout is one setting per tool. What changes is underneath: every request is trimmed, routed, cached or batched to cost less, and has a name, a budget and a log line.

Nothing to install on anyone's laptop

Proxen is an address and a key. There is no local daemon, no browser extension and no wrapper binary to push through device management. The tool an engineer used yesterday is the tool they use today. If a tool ever needs to be pointed straight at a provider again, that is the same one setting changed back.

Same answers, checked

Before a rule goes live we test it on your own work and compare the answers with what your team gets today. Hard tasks keep the strongest model. Every request runs on provider accounts your company owns, and the log shows which model answered, so engineers can check for themselves.

The provider key never leaves the server

Engineers and agents hold a Proxen key, not an Anthropic, OpenAI or Google key. If a laptop is lost or a key is pasted into a repo, the admin revokes that one key and the damage stops at its budget. The provider keys stay in Proxen, where nobody in the company can read them.

1

Setting to change in most tools: the base URL

20

Tools, SDKs and frameworks above that can point at a custom address

$
13

Average Claude Code cost per developer per active day (Anthropic, Claude Code docs)

Further reading

Cutting the AI bill without cutting usage

Blog

Pointing Claude Code, Codex and Cursor at one gateway

Blog

A key per person: how per-person AI keys work

Blog

Offboarding someone who knows the AI password

Blog

Keep your tools. Change one address. Pay less.

Send us one month of usage, or tell us which tools and agents you run. We show you where their cost leaks and what the proxy would save. Free.

By sending this you agree to Proxen's Terms and Privacy Policy.