🜂

vaked inference node

a sovereign CPU inference node in the entheai / vaked constellation

status flowing
modelqwen3-coder:30b
quantizationQ4 · ~18 GB · 30B/3B-active
computeCPU · ~19 tokens/sec
apiOpenAI-compatible

Quickstart

1 · Free, no key — point any OpenAI-compatible client at https://coder.vaked.dev/v1:

curl https://coder.vaked.dev/v1/chat/completions \
  -d '{"model":"qwen3-coder:30b","messages":[{"role":"user","content":"quicksort in rust"}]}'

2 · Go fast (paid)subscribe (€10), get your vaked_sk_… key, and hit the GPU-backed model:

curl https://coder.vaked.dev/v1/premium/chat/completions \
  -H "Authorization: Bearer vaked_sk_YOUR_KEY" \
  -d '{"model":"qwen3-coder-next","messages":[{"role":"user","content":"refactor this"}]}'

Free = CPU, ~19 tok/s, shared. Paid = Qwen3-Coder-Next on an H200, ~60+ tok/s. Full usage guide.

What this is

An honest, public, open-model inference endpoint. The free tier runs a quantized Qwen3-Coder-30B (MoE, only 3B active) on plain CPU — ~19 tok/s, no GPU; paid tiers route to an H200. It is part of a small constellation of nodes that generate and serve open knowledge. The free tier is public and unauthenticated — use it kindly; it is one machine, shared.

Use it

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."}]
  }'

Drop-in for any OpenAI client — set the base URL to https://178-156-242-147.sslip.io/v1 and the model above. No key required.

Tiers

freerate-limited · no key · conversations feed the public corpus (opt-out)
€10/mo · unlimitedfair-use unlimited · private by default
€20/mo · prioritypriority queue · premium model on request

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

Prices are VAT-inclusive — you pay exactly €10 or €20. After checkout you're handed an API key. Opt in to the public dogfeed corpus from your account page for −€3/mo. Details: usage guide · privacy.

Why it exists

Because a model you can point at, read the weights of, and run on hardware you hold is different from one you rent. This one is small and slow and owned. The larger argument lives at vaked.dev and entheai.