OpenAI Compatible API Key: Get One and Send a Call
An OpenAI compatible API key is a normal bearer token that works with any OpenAI SDK once you point the client at a different base URL. This page shows how to get one for the hosted text API on this site and make a working call in a few minutes.
UpdatedSpeech To Text API

What the key actually is
"OpenAI compatible" describes the shape of the HTTP API, not the company behind it. The service accepts the same JSON body on POST /v1/chat/completions that the OpenAI SDKs send, and it reads the same Authorization: Bearer header. Because of that, a key from this service slots into any tool that lets you set a custom base URL and API key: the official Python and Node libraries, LangChain, plain curl, and most desktop chat clients.
The key is not an OpenAI key, and it will not work on api.openai.com. The reverse is also true: an OpenAI key will not work here. Each key belongs to one service, and this one runs a single open-weight model, tuned to answer without content refusals for lawful adult use, hosted by an independent provider. It is not the official API of any model vendor and it does not route to other vendors.
Here is what you need to remember. The base URL for this site is https://api.speechtotextapis.com/v1. The model ID is uncensored. The key is whatever string you get after signing up. Everything below is a variation on those three values.
The service is text only. It handles chat completions, streaming, tool calling and JSON mode. It does not generate images or audio, it does not offer embeddings, and it does not fine-tune models. If you came here from a speech pipeline, the typical job is the text step after transcription: cleaning a transcript, splitting it into scenes, rewriting dialogue, or extracting structured data. The voice script processing guide covers that side in more detail.
Get your key in four steps
- Sign up. Use Google, or register with an email address and a password.
- Open the API key page. It lives at the key page on this site. Your key is created and shown there.
- Claim the trial credit. A button on the same page gives you $0.50 of trial credit with one click. It is one per person, valid for 7 days, and it needs no card.
- Copy the key into a secret store. An environment variable is enough for a local test. For anything deployed, use your host's secret manager.
The trial is meant to prove that the integration works before you pay anything. At the current price it is enough for a lot of short requests, and the cost section further down shows the arithmetic. When you are ready to keep going, the pricing page lists the top-up amounts. Credit is prepaid, there is no subscription, and the credit never expires, so there is no pressure to spend a top-up before some deadline.
Trial keys are limited more tightly than full keys: 2 concurrent requests and 60 requests per minute. That is plenty for a first script. If you plan to fan out dozens of transcript chunks in parallel, expect to hit the limit until you move to a funded account.
The three settings every client asks for
Whatever software you use, the setup screen asks for the same three things under slightly different names.
| Setting | Typical field name | Value |
|---|---|---|
| Endpoint | Base URL, API base, host | https://api.speechtotextapis.com/v1 |
| Credential | API key, token, secret | Your key from the key page |
| Model | Model, model name, engine | uncensored |
One detail trips people up: the base URL already ends in /v1. The SDKs append /chat/completions themselves, so you should not add it. If your client asks for the full URL of the chat endpoint instead of a base, the value is https://api.speechtotextapis.com/v1/chat/completions.
If a tool only offers an "OpenAI" preset with a hard-coded host and no way to edit it, it cannot talk to this service. Look for a "custom", "OpenAI-compatible" or "advanced" option that exposes the base URL field.
Your first request
Start with curl, because it removes every layer between you and the API. Replace YOUR_KEY with the key you copied.
curl https://api.speechtotextapis.com/v1/chat/completions \
-H "Authorization: Bearer YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "uncensored",
"messages": [
{"role": "system", "content": "You clean up raw transcripts. Keep the meaning, fix punctuation."},
{"role": "user", "content": "so um we shipped the the beta on tuesday and uh nobody complained"}
]
}'
A successful reply is a JSON object with a choices array. The text is in choices[0].message.content, and a usage block reports the tokens you were billed for.
The same call with the official Python library looks like this. Only base_url and api_key differ from a stock OpenAI setup.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.speechtotextapis.com/v1",
api_key=os.environ["API_KEY"],
)
resp = client.chat.completions.create(
model="uncensored",
messages=[
{"role": "system", "content": "You clean up raw transcripts. Keep the meaning, fix punctuation."},
{"role": "user", "content": "so um we shipped the the beta on tuesday and uh nobody complained"},
],
)
print(resp.choices[0].message.content)
And in Node, using the same package:
import OpenAI from "openai";
const client = new OpenAI({
baseURL: "https://api.speechtotextapis.com/v1",
apiKey: process.env.API_KEY,
});
const resp = await client.chat.completions.create({
model: "uncensored",
messages: [
{ role: "user", content: "Rewrite this in plain English: the beta shipped on Tuesday." },
],
});
console.log(resp.choices[0].message.content);
You do not need to set max_tokens. When it is omitted, the completion can use whatever is left of the 100,000-token window after your prompt. Set it yourself if you want a hard ceiling on cost per call.
Check that the key works without spending tokens
Before wiring a key into a larger system, make one cheap call to the models endpoint. It confirms the key is accepted and the base URL is right.
curl -i https://api.speechtotextapis.com/v1/models \
-H "Authorization: Bearer YOUR_KEY"
The -i flag prints response headers, which is where this API reports useful state:
X-Request-Ididentifies the request. Include it if you contact support about a specific call.X-Balance-USDshows the remaining credit on the account.X-RateLimit-Limit-RequestsandX-RateLimit-Limit-Concurrencyshow the limits that apply to your key, so you can tell at a glance if the key has trial or full limits.
Reading X-Balance-USD in your own code is a cheap way to alert yourself before credit runs out. Log it once per job rather than once per request if you do not want noisy output.
What a key costs to use
Having a key costs nothing. You pay for tokens: $0.15 per 1M input tokens and $0.50 per 1M output tokens, drawn from prepaid credit. Failed or refused requests are not charged.
A worked example with the transcript cleanup call above. Say a prompt is 1,000 tokens and the cleaned reply is 500 tokens.
- Input: 1,000 / 1,000,000 x $0.15 = $0.00015
- Output: 500 / 1,000,000 x $0.50 = $0.00025
- Total: $0.0004 per call
The $0.50 trial therefore covers about 1,250 calls of that size ($0.50 / $0.0004). Your own numbers will differ, because transcripts vary in length, but the method is the same: count the tokens in, count the tokens out, multiply.
When you top up, any whole amount from $20 to $500 is accepted. Larger top-ups carry a bonus: +5% from $50, +10% from $100 and +15% from $200. A $100 top-up, for example, gives you $110 of credit. You can pay with crypto (USDT on TRC20, USDC on Base).
Keeping the key safe
An API key is a password for your balance. Anyone who has it can spend your credit, so treat it that way.
- Keep it out of source control. Read it from an environment variable or a secret manager.
- Do not ship it in browser JavaScript or a mobile app bundle. Anything that runs on the user's device can be inspected. Put a small server between your front end and the API and let the server add the header.
- Do not paste it into screenshots, tickets or chat messages.
- If you think a key has leaked, stop using it and contact support right away rather than waiting to see if anything happens.
On privacy: prompts sent to the API are not used for training. That is a statement about the service, not a reason to send secrets you do not need to send. Redact personal data from transcripts when the task does not require it.
When a call fails
Most first-day problems come from one of four causes.
The base URL is wrong
The two usual mistakes are a missing /v1 and a doubled /chat/completions. Compare your value character by character with the one in the table above.
The key has a stray character
Copying from a page often brings along a trailing space or newline. Trim the value in your code, and check that the header reads Bearer, one space, then the key.
The model name is not uncensored
Clients often default to another model name. Set the model explicitly on every call.
HTTP 429
There are two flavours, and the error code tells them apart. rate_limited means you sent too many requests per minute. concurrency means too many were in flight at once. A streamed request holds its concurrency slot until the stream ends, so a slow reader can cause this error even at a low request rate. A 429 with a Retry-After: 5 header means the service is briefly overloaded; wait the five seconds and retry. A short exponential backoff in your client handles all three cases.
For the full parameter list, streaming format and tool calling details, see the API documentation.
More guides
Questions and answers
Is this the same as an OpenAI key?
No. It follows the same request format, so OpenAI SDKs work with it, but it is issued by this service and only works against this service's base URL.
Do I need a credit card to get a key?
No. Sign up with Google or email, then claim the $0.50 trial credit with one click on the key page. The trial needs no card and is valid for 7 days.
Which model name should I use?
Use uncensored in the model field. A call to GET /v1/models will list what the key can access.
Can I use the same key from several servers?
Yes, but the limits apply to the key as a whole: 8 concurrent requests and 300 requests per minute on a full key, 2 and 60 on a trial key.
Does the key work for images or speech?
No. The API is text only. It does not generate or transcribe audio or images, and it has no embeddings or fine-tuning.
Your key is one form away
Create an account, copy the key, change the base URL. That is the whole setup.