Questions, answered plainly.

No. Prompts and outputs are never written to logs, disk or the database; finished videos are the one exception, kept encrypted for an hour. A request leaves only its usage (key, model, tokens or units, and cost), merged into hourly, then daily, then weekly totals.

Copy link How a request flows →

The database runs in Quebec, on hardware we own or rent and run ourselves. Some models run on GPUs we rent; others go to providers with a no-logs policy. The team works from Quebec and Tennessee.

Copy link

Yes. Sign up with only a username, top up with $5 slips that can't be traced back to your payment, and use the site and API over Tor. Your requests reach the models without your IP address, your headers or the ID fields SDKs add.

Copy link How anonymity works →

Point Aider at the OpenAI-compatible base URL:

export OPENAI_API_BASE=https://api.yourbrand.sh/v1
export OPENAI_API_KEY=$YOURBRAND_KEY
aider --model openai/qwen3.8-27b-abliterated

Copy link Agent setup docs →

Yes, both, in the same format the OpenAI and Anthropic SDKs already expect.

Copy link Streaming docs →

They're set by concurrent requests, not requests per minute: 2 on pay as you go, 3 on Starter, 5 on Pro and 8 on Max. A request over the limit gets a 429 with a Retry-After header.

Copy link Rate limits →

Yes, through Etheroute.io: Bitcoin, Ethereum and its L2s like Base and Arbitrum, Solana, Litecoin, USDC, USDT and more. Card, PayPal and Apple Pay go through our payment processor. Either way, credit comes as $5 slips, and nothing is refunded.

Copy link

No. Slips never expire, and plan credit is added to your balance like any other credit.

Copy link

Nothing that's legal. Requests and outputs are checked only for what the law forbids, such as child sexual abuse material. The models have their blanket refusals taken out, so they'll write dark fiction, explain how an exploit works, and answer medical, legal and drug questions plainly. What closes an account is in the acceptable use policy: content that sexualises minors, threats against real people, malware against systems you don't own, and fraud.

Copy link Acceptable use →

qwen3.8-27b-abliterated for coding agents, since it has the most reliable tool calls. llama4.5-70b-abliterated for long documents, once it leaves the queue. For search, speech, images or video, pick by task on the models page.

Copy link Compare models →