Quickstart

1. Deploy into your account

The proxy runs in your own Vercel project, pointed at your upstream base URL and key. Your provider key stays an environment variable in your project — we never receive it. The deploy and its environment variables are covered in the repository docs.

2. Issue each client a capped key

Client keys are stored only as SHA-256 hashes. Generate a key, hash it, and give it a monthly USD budget and an RPM limit:

node -e "console.log(require('crypto').createHash('sha256').update('sk-acme-...').digest('hex'))"

The plaintext key is shown once, when it is issued. Over budget returns 402, over rate returns 429, an unknown key returns 401.

3. Point your clients at /v1

Continue

Add an OpenAI-compatible model to ~/.continue/config.yaml (or config.json on older versions):

name: streamproof
version: 0.0.1
models:
  - name: streamproof
    provider: openai
    model: gpt-4o-mini
    apiBase: https://YOUR-DEPLOYMENT/v1
    apiKey: ${{ secrets.STREAMPROOF_API_KEY }}
    roles:
      - chat
      - edit

apiBase must include the /v1 suffix. Continue appends /chat/completions to it.

Cline (VS Code)

In the Cline panel, choose OpenAI Compatible and fill in:

Cline appends /chat/completions, so the base URL includes /v1.

Verify with curl

curl -sS https://YOUR-DEPLOYMENT/v1/chat/completions \
  -H "authorization: Bearer $STREAMPROOF_API_KEY" \
  -H "content-type: application/json" \
  -d '{"model":"gpt-4o-mini","messages":[{"role":"user","content":"ping"}],"stream":false}'

A working response is HTTP 200 with an OpenAI-shaped body and choices[0].message.content populated. For streaming, add "stream": true and an accept: text/event-stream header.

When something goes wrong