sealrelay.com
OpenAI-compatible · Pay per token

No filters.
No guardrails.
No apologies.

Sanitizing a model’s output is patronizing, unhealthy, and a colossal waste of human effort. So we don’t.

You’re an adult. Prompt responsibly.

[01] Plugs into whatever you’re already using

  • OpenCode
  • pi
  • Warp
  • Aider
  • Cline
  • Cursor
  • Continue
  • Zed
[02] Setup

Base URL, API key, done.

Point the OpenAI client you already use at us. Chat completions, streaming and tool calls work as expected.

→ Docs
terminal
$ curl https://api.sealrelay.com/v1/chat/completions \
  -H "Authorization: Bearer $API_KEY" \
  -d '{
    "model": "qwen3.8-27b-uncensored",
    "messages": [{"role": "user", "content": "..."}]
  }'
[03] Models & pricing

Pay per token. No subscription.

Prices are per million tokens. One balance, any model.

DefaultLive

The main model. Strong reasoning and code, long context, and nothing stripped out of the answer. If you only try one, try this one.

Context 256K · Model id: qwen3.8-27b-uncensored

Input$0.30 / M
Output$2.50 / M
Use this model
Incoming

GLM 5.3 Flash Uncensored

Fast and cheap. Built for agent loops, tool calls and anything where latency matters more than depth.

Pricing announced at launch
Incoming

DeepSeek V4.1

Heavy reasoning for the hard problems. Slower, pricier, worth it when the answer has to be right.

Pricing announced at launch
Incoming

Qwen 3.8 Next

The next flagship. Same treatment as the 27B, more of everything. Landing soon.

Pricing announced at launch
[04] FAQ

The awkward questions.

How do I get credit?
With an invite code, during early access. Enter it when you sign up, or later on the Billing page of your dashboard; it adds its credit to your balance once. Every request deducts from the balance at the per-token prices above, and when it hits zero, requests stop.
Can I pay instead?
Soon. Top-ups open after early access, prepaid in crypto: BTC, ETH, USDC, USDT and SOL, on Bitcoin, Ethereum, Base, Arbitrum, Optimism, Solana and Tron. No card, no bank, no fiat on-ramp.
What if I lose my password?
Use the recovery code you saved when you signed up: Log in → Forgot password? takes your username, the code and a new password. Lose the code too and the account is gone, along with its balance. That’s the trade: we can’t recover what we never knew. A new code can be made in Settings at any time.
Is the API OpenAI-compatible?
Yes. Change the base URL, set the API key, pick a model id. Chat completions, streaming and tool calls work the way your existing client expects.
What does “uncensored” mean here?
The models answer the prompt. No refusal training, no output filter, no moralizing preamble. What you do with that is on you — see the top of the page.
Do you keep logs?
Not of prompts or outputs. Each API request leaves one line for 7 days: time, model, status, duration and a random request ID, with no key, account, IP or content. Usage is kept as daily token totals per key, for billing, for 90 days. Site visits are counted as daily totals, without cookies. The exception: if a model server crashes, its error log can include the token IDs of the requests it was running. The full list is on the privacy page.
[05] Accounts

Username. Password.
That’s the form.

Accounts are a username and a password. Credit comes from an invite code for now; crypto top-ups open soon. No email, not even an optional one. A recovery code, shown once at signup, resets a lost password.

Create an account