One shared API key. Zero spending limit.

Streamproof is a drop-in proxy for any OpenAI-compatible model. Each client gets their own key and a hard monthly budget, and broken model replies get repaired on the way through.

See how it works Try the budget cap

Works with Continue, Cline, the OpenAI SDKs and anything that lets you change the base URL.

402 budget reached

One client’s runaway agent, overnight

$0.00

Shared keyno limit

Streamproofcap $50

What it actually does

You change one line, the base URL. From then on every request to your models passes through five checks that you would otherwise have to build and maintain yourself.

IdentifyEach client sends their own key. We store only its hash, so a leak of the config exposes nothing usable.

LimitA monthly dollar budget and a requests-per-minute cap. Over budget gets a 402, too fast gets a 429.

CleanParameters that make some models fail, like min_p, are removed before the request goes out.

RepairThe streamed reply is fixed live, so editors and apps stop showing empty answers.

BillToken usage times your price list is added to that client’s spend, ready for invoicing.

Why replies come back empty

Several fast open models send usage statistics on every streamed chunk. Tools like Continue read that as the end of the message and drop everything after it. Streamproof removes the noise and keeps the text and tool calls.

Direct to the model empty reply

Through Streamproof delivered

Try the budget cap

This is the same logic the proxy runs. Each request costs $6.40. When acme reaches $50, the next one is refused.

No surprise invoicesThe limit is enforced before the request reaches your provider.

Per client, not per accountOne heavy client cannot use up the budget of another.

You set the pricesAny model, any provider, with your own per-million-token rates.

acme$0.00 / $50

globex$44.9

initech$8.5

umbrella$3.1

Change one line.

Point your client at your Streamproof address and give each customer their own key. Nothing else in your code changes.

// before
baseURL: "https://integrate.api.nvidia.com/v1"
apiKey:  process.env.SHARED_KEY

// after
baseURL: "https://ai.your-agency.com/api"
apiKey:  "sk-acme-…"  // acme's own key and budget

Built for people who resell AI.

If other people’s usage lands on your invoice, this is for you.

AI agencies and freelancersRun models for several clients from one account and bill each one fairly.

Small dev teamsGive every developer or project a cap so one loop cannot burn the month.

Open-model usersUse Kimi, Step or Nemotron in Continue without silent empty answers.

Without it, and next to the alternatives.

Per-client budgets are not unique to us. OpenRouter, LiteLLM and Vercel’s own gateway have them. What differs is what else you get, and how much you have to run yourself.

Without Streamproof

One shared provider key, so one bad loop is billed to you
You learn about overspend from the invoice
Open models answer, but Continue shows an empty reply
You split costs between clients by hand in a spreadsheet

With Streamproof

Each client has their own key and a dollar cap that is enforced before the call
The next request after the cap gets a clear 402
Streams are repaired, so replies arrive in your editor
Spend per client is one authenticated API call away

	Shared key	OpenRouter	LiteLLM, self-hosted	Vercel AI Gateway	Streamproof
Dollar budget per client	No	Yes	Yes, needs a database	Yes	Yes
Repairs empty replies from open models	No	Not its purpose	Not its purpose	Not its purpose	Yes
Caps tokens per request	No	Provider settings	Yes	Provider settings	Yes, per client
Server you must run	None	None, hosted by them	Yes, plus Postgres	None, hosted by Vercel	None, runs in your Vercel account
Provider keys stay with you	Yes	Via their credits	Yes	Via their gateway	Yes
Installed and tested for you	-	No	No	No	Yes, €450 once

Based on each vendor’s public documentation. Check current terms before you rely on a row. If budgets are all you need, OpenRouter or LiteLLM are good choices. Streamproof fits when you also run open models in editors like Continue and want it installed for you.

We set it up on your Vercel account.

You keep your provider keys and your data. We configure models, prices and clients, test with your tools, then hand it over.

Setup

€450

One time

Deployed to your Vercel project
Up to 5 clients and 3 models
Tested with Continue or Cline

Book a setup

Managed

€49 /month

Added to any setup

New models and price updates
Spend alerts per client
Fixes when providers change

Add managed