usage guide

how to use the vaked inference node · home · privacy

Quick start (free, no key)

The node speaks the OpenAI API. Point any client at the base URL and go — no account, no key for the free tier.

curl https://178-156-242-147.sslip.io/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model":"qwen3-coder:30b","messages":[{"role":"user","content":"Write quicksort in Rust."}]}'

Base URL: https://coder.vaked.dev/v1 — your request runs on the H200 (Qwen3-Coder-Next), same as paid; the model name in the body is ignored on the free path.

Tiers

	Free	€10/mo · Unlimited	€20/mo · Priority
rate	rate-limited	fair-use unlimited	fair-use unlimited + priority queue
auth	none / IP	API key	API key
model	Qwen3-Coder-30B · CPU	Qwen3-Coder-Next-80B · H200	Qwen3-Coder-Next-80B · H200 + priority
your data	→ public corpus (opt-out)	private (opt-in for −€3)	private (opt-in for −€3)

Subscribe · €10 Unlimited → Subscribe · €20 Priority →

Prices are VAT-inclusive — you pay exactly €10 or €20. EU businesses: enter your VAT ID at checkout for reverse-charge (no VAT added).

Getting & using your key

Subscribe via a link above → complete Stripe checkout with any card → you're redirected to your account page, which shows your API key (format vaked_sk_…). Treat it like a password. Then send it as a bearer token:

curl https://178-156-242-147.sslip.io/v1/chat/completions \
  -H "Authorization: Bearer vaked_sk_YOUR_KEY" \
  -d '{"model":"qwen3-coder:30b","messages":[{"role":"user","content":"refactor this"}]}'

Premium (Priority tier) routes under /v1/premium/*. A non-Priority key there returns 403.

Request schema (OpenAPI)

Copy-pastable OpenAPI 3.1 for the chat endpoint:

{
  "openapi": "3.1.0",
  "info": { "title": "Vaked Coding Inference", "version": "1.0.0" },
  "servers": [{ "url": "https://coder.vaked.dev/v1" }],
  "paths": {
    "/chat/completions": {
      "post": {
        "summary": "Chat completion — free tier (no key). Paid: /premium/chat/completions with Bearer vaked_sk_…",
        "requestBody": { "required": true, "content": { "application/json": { "schema": {
          "type": "object",
          "required": ["messages"],
          "properties": {
            "model":       { "type": "string",  "example": "qwen3-coder" },
            "messages":    { "type": "array", "items": { "type": "object",
              "required": ["role","content"],
              "properties": {
                "role":    { "type": "string", "enum": ["system","user","assistant"] },
                "content": { "type": "string" } } } },
            "max_tokens":  { "type": "integer", "default": 512 },
            "temperature": { "type": "number",  "default": 0.7 },
            "stream":      { "type": "boolean", "default": false }
          } } } } },
        "responses": { "200": { "description": "OpenAI-compatible chat completion" } }
      }
    }
  }
}

Dogfeed — opt out (free) / opt in (paid)

The corpus is ultrawhale-dogfood, and the generator is open source: github.com/peterlodri-sec/ultrameshai.

Honest state: right now the corpus is fed by our own generation loop (the model producing coding Q&A on seed topics) — not by tapping your traffic. So free conversations aren't being published yet. When free-traffic capture goes live, opt out per request:

curl https://coder.vaked.dev/v1/chat/completions \
  -H 'X-Vaked-Dogfeed: off' \
  -d '{"messages":[{"role":"user","content":"…"}]}'

Paid opt-in (private by default): after subscribing you're redirected to your account page — the dogfeed checkbox there controls whether your conversations contribute to the corpus (opting in earns −€3/mo).

Models

Fair use & the dogfeed bargain

"Unlimited" means fair-use unlimited — full speed for normal heavy use; the abusive extreme is throttled, never cut. The honest bargain, stated plainly:

• Free — your conversations, with personal information scrubbed out, become open training data. You pay with data. (You can opt out.)
• Paid — private by default. If you choose to contribute to the corpus, you get −€3/mo, because you're feeding what makes the model cheaper.

PII scrubbing is best-effort — never paste live secrets, tokens, or credentials into any prompt on the free tier.

Free vs paid — what actually differs

It's a model ladder plus terms:

• Model — free runs Qwen3-Coder-30B (a real dedicated coder); paid upgrades to Qwen3-Coder-Next-80B on an H200 (bigger, faster).
• Privacy — free conversations feed the public corpus (opt-out); paid are private (opt-in for −€3).
• Priority & limits — free is rate-limited; paid is fair-use-unlimited, and Priority (€20) jumps the queue.

Honest state

Live today: free = Qwen3-Coder-30B on CPU (~19 tok/s, keyless, with a 65s shared Redis cache); paid = Qwen3-Coder-Next-80B on an H200 (~60 tok/s) at /v1/premium. Honest caveats: one H200 serves one model, so a bigger "best" model for Priority (e.g. Qwen3-Coder-480B) and putting the free 30B on its own GPU both need more GPUs — not provisioned yet. The per-tier rate limiter and usage-based autoscaling aren't built yet either. This page will always say what is real. 🜂