Updated

Uncensored LLM API docs

This reference covers what an uncensored or abliterated LLM API exposes, with Silk Compute's API as the worked example. Each endpoint page lists its parameters and shows the call in several languages.

https://api.silkcompute.com
# Either header works
curl https://api.silkcompute.com/models \
  -H "Authorization: Bearer $SILK_API_KEY"

curl https://api.silkcompute.com/models \
  -H "x-api-key: $SILK_API_KEY"
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://api.silkcompute.com",
    api_key=os.environ["SILK_API_KEY"],
)
import OpenAI from 'openai'

const client = new OpenAI({
  baseURL: 'https://api.silkcompute.com',
  apiKey: process.env.SILK_API_KEY,
})
curl https://api.silkcompute.com/balance \
  -H "Authorization: Bearer $SILK_API_KEY"
import os, urllib.request

req = urllib.request.Request(
    "https://api.silkcompute.com/balance",
    headers={"Authorization": f"Bearer {os.environ['SILK_API_KEY']}"},
)
print(urllib.request.urlopen(req).read().decode())
const res = await fetch('https://api.silkcompute.com/balance', {
  headers: { Authorization: 'Bearer ' + process.env.SILK_API_KEY },
})
console.log(await res.json())

How it works

Base URLhttps://api.silkcompute.comTools that want /v1 at the end can use https://api.silkcompute.com/v1.
AuthenticationBearer or x-api-keySend the key in either header. Requests without one return 401 invalid_api_key.
Limits8 requests at onceA ninth waits up to 5 seconds, then gets 429. Bodies go up to 64 MiB, and every model has a 1,048,576-token context.

Endpoints

GET /balance returns the account's remaining credit. Billing is per token, from prepaid credits or a monthly plan.

EndpointPath
Chat CompletionsPOST /chat/completions
MessagesPOST /v1/messages
ResponsesPOST /responses
List modelsGET /models
BalanceGET /balance

Formats compared

All three reach the same models, prices and limits.

FeatureChat CompletionsReferenceResponsesReferenceMessagesReference
Endpoint/chat/completions/responses/v1/messages
Streaming
Tool calling
Images
VideoDropped, still billedNo, returns 400
Reasoning effort
Thinking textWith thinking on
Prompt caching
previous_response_idNot applicableNot applicable
PDFs and token countingNot applicableNot applicable
Hosted tools such as web search