Open-model inference, billed per token in USDC on Base. No subscription. No signup. No logs. Just an API key and your wallet.
No card. No email. Pay only for the tokens you use.
from openai import OpenAI
client = OpenAI(
base_url="https://voltdapp.site/api/v1",
api_key="volt_sk_..."
)
client.chat.completions.create(
model="llama-3.3-70b",
messages=[{"role": "user", "content": "hey"}]
)Coinbase Wallet, MetaMask, Rabby — anything on Base. One free signature, no password, no email.
Sign an EIP-3009 authorization. We pay the gas, so you never need ETH. Funds settle on Base in seconds.
Change one base URL. Any OpenAI-compatible SDK just works — streaming and tool calls included.
What you see is what you're charged. No tiers, no seats, no surprise invoice.
| Model | Context | Input /1M | Output /1M |
|---|---|---|---|
Llama 3.3 70B llama-3.3-70b Best all-rounder. Open weights. | 128k | $0.46 | $0.80 |
Llama 3.1 8B llama-3.1-8b Cheap and fast. Great for bulk jobs. | 128k | $0.06 | $0.10 |
Qwen 2.5 72B qwen-2.5-72b Strong at code and reasoning. | 32k | $0.46 | $0.80 |
DeepSeek V3 deepseek-v3 Frontier-class open model. | 64k | $0.98 | $1.78 |
Mistral Small 24B mistral-small Balanced. Good default for chat. | 32k | $0.10 | $0.16 |
Open-weight models only. We don't resell GPT or Claude — that breaks their terms and it isn't a business you can build on.
Most providers keep your prompts for 30 days “for safety.” We keep a token count and a timestamp.
We store token counts and a timestamp. Not your prompts, not the completions. Nothing to leak, nothing to subpoena.
A wallet address is your whole account. No email, no name, no phone, no KYC form.
Requests reach the GPU provider stripped of your key, your IP, and your headers. They see a prompt from us, not from you.
OpenAI-compatible in, OpenAI-compatible out. Swapping away is a one-line change. No lock-in by design.
Today your prompt is decrypted in memory on a GPU we rent, so “we don't log” is a policy we hold ourselves to — not yet something maths enforces. Hardware-attested TEEs (Intel TDX, AMD SEV-SNP, NVIDIA CC) make it enforceable, and that's on the roadmap. Anyone selling you “unbreakable” private inference today is overselling it.
Connect a wallet, grab a key, send your first request. Takes about a minute.
Get your API key →You pay per token, listed below, in USDC. No subscription and no monthly minimum. Top up from 1 USDC and spend it whenever — credits do not expire.
No. You sign an EIP-3009 authorization off-chain and we submit it, paying the gas ourselves. You only ever need USDC on Base.
Open-weight models only — Llama, Qwen, DeepSeek, Mistral. We deliberately do not resell GPT or Claude: reselling closed-model API access breaks those providers’ terms, and a business built on that gets switched off overnight.
No. We do not train on anything, and we do not retain prompts to train on later. We keep token counts for billing and that is the whole record.
Requests route to a fallback provider automatically. If every provider is unreachable your reservation is refunded in full — a failed request never costs you anything.