Quickstart

The API follows OpenAI's, so existing clients work. Set the base URL to https://api.abliterate.tech/v1 and use an inference key.

1. Create an account and add at least $5 of credit in the console.
2. Make an inference key under API keys. It's shown once, so store it safely.
3. Send a request.

from openai import OpenAI

client = OpenAI(
    base_url="https://api.abliterate.tech/v1",
    api_key="sk-abl-...",
    timeout=600,  # the first request on an idle model starts a GPU
)

stream = client.chat.completions.create(
    model="qwen3-1.7b-abliterated",
    messages=[{"role": "user", "content": "Write a port scanner in Go."}],
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://api.abliterate.tech/v1",
  apiKey: process.env.ABLITERATE_API_KEY,
  timeout: 600_000,
});

const stream = await client.chat.completions.create({
  model: "qwen3-1.7b-abliterated",
  messages: [{ role: "user", content: "Write a port scanner in Go." }],
  stream: true,
});
for await (const chunk of stream) {
  process.stdout.write(chunk.choices[0]?.delta?.content ?? "");
}

curl https://api.abliterate.tech/v1/chat/completions \
  -H "Authorization: Bearer $ABLITERATE_API_KEY" \
  -H "Content-Type: application/json" \
  -i -d '{
    "model": "qwen3-1.7b-abliterated",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Anthropic format

The same keys work with Anthropic's SDK, whose base URL leaves off /v1.

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.abliterate.tech",
    api_key="sk-abl-...",
    timeout=600,
)
message = client.messages.create(
    model="qwen3-1.7b-abliterated",
    max_tokens=256,
    messages=[{"role": "user", "content": "Hello"}],
)
print(message.content[0].text)

Starting a model

When no GPU worker is running, the first request has to start one, which usually takes a few minutes. The request waits rather than failing, and streams send a comment every 15 seconds so that nothing times out.

• Give your client a timeout of at least 5 minutes.
• Don't retry early. A request you've abandoned is cancelled, and you won't pay for what never ran.
• GET /v1/models/{id} gives a hint in state: warm if a worker looks to be up, cold if your next request will probably start one.

Keys

Put inference keys in your applications and keep admin keys out of them, so a leaked key can't reach your card. A key can have an expiry date and its own monthly spend limit, and can belong to a project.

Limits

New accounts can send 120 requests a minute and 8 at once per key. Limits rise as you pay in more in total, at $20, $50 and $200. GET /v1/account shows yours.