Pre-release / diagnostic mode

OPEN-MODEL INFRASTRUCTURE · 001

The modelshouldn't bea black box.

BrokenGPT is a less-restrictive chat and API surface for open models—built around visible model identity, measured limits, and rules you can actually read.

Enter the demo Read the API docs

The default build returns a diagnostic response until an inference provider is connected.

OPENAI-COMPATIBLE REQUEST SHAPE◆STREAMING RESPONSES◆POSTGRES METERING◆DISCLOSED MODEL IDENTITY

Demo model status

The pipes work. The engine stays labelled.

BrokenGPT separates the platform from the model provider, so the account, chat, streaming, API, quotas, and usage path can be verified before production inference is attached.

Default deployment

DIAGNOSTICNot a production language model

Gateway path

Chat and API traffic run through the same metered request path.

READY TO TEST

Inference provider

Connect an OpenAI-compatible provider or a self-hosted vLLM endpoint.

AWAITING CONFIG

Model identity

Upstream model, context, and deployment notes belong in the public disclosure.

REQUIRED FOR LAUNCH

This status describes the default repository configuration, not a synthetic uptime claim. Verify your deployed environment before presenting it as production-ready.

For builders

One request shape. No proprietary fog.

Use familiar chat-completions semantics, bearer API keys, server-sent events, and explicit quota headers. The docs define the supported subset.

THE CONTRACT

Change the base URL. Keep your message structure.

The public model alias stays stable while the actual upstream model is disclosed separately. That keeps client integration simple without hiding what serves the response.

Bearer-key auth with revocable hashed API keys.
JSON or SSE using the same completion endpoint.
Visible limits through request IDs and rate headers.

Open integration guide

POST /v1/chat/completions

cURL

curl "$BROKENGPT_BASE_URL/v1/chat/completions" \
  -H "Authorization: Bearer $BROKENGPT_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "broken-one",
    "messages": [
      { "role": "user", "content": "Explain the tradeoff plainly." }
    ],
    "stream": true,
    "max_tokens": 800
  }'

AUTHBEARER

STREAMSSE

FORMATJSON

Transparent by construction

Less filtered. Still accountable.

‘Uncensored’ is not a magic property. We would rather show you the model, the service boundary, and the policy than sell a vague promise.

Identity before hype

Publish the upstream model, context window, serving mode, and known limitations before calling it live.

Read model disclosure

Limits in numbers

Monthly tokens, throughput, and maximum output are specific per plan. No “unlimited” asterisks.

Compare plans

Freedom with a perimeter

Broad expression is compatible with a firm boundary against illegal abuse, exploitation, and attacks.

Read acceptable use

Pay as you go

Pay for the tokens you use, nothing else.

Add credits, spend them per token across chat and the API. No subscription, no monthly reset — and $1 free to start.

Pay as you go

$0.50/ 1M input tokens

Add credits, spend them per token. No subscription, no monthly reset.

$2.00 / 1M output tokens
Top up any amount — credits never expire
Chat + OpenAI-compatible API on one balance
$1 in free credit when you sign up

Create free account

02 / TEAMS

Enterprise

$20/ seat / month

Seat-based plan for teams that want higher throughput and priority.

Higher rate limits + priority routing
Team seats, managed in the billing portal
Token usage billed to your shared credit balance
Volume pricing on request

Start Enterprise

$0.50 / 1M input tokens · $2.00 / 1M output tokens · credits never expire.

See the full breakdown

Straight answers

Before you wire it into anything.

What does ‘uncensored’ mean here?

We use narrower, testable language: BrokenGPT is being built for less-restrictive access to disclosed open models. Every model still has behavior inherited from its training, and the service still enforces an acceptable-use policy for illegal abuse and attacks on people or systems.

Is a production language model already connected?

Not in the default deployment. It starts with a clearly labelled diagnostic responder so chat, streaming, API keys, quotas, and usage tracking can be tested. The model disclosure identifies the upstream model before a production endpoint goes live.

Can I use an OpenAI SDK?

The chat-completions endpoint follows an OpenAI-compatible request shape for messages, sampling controls, token limits, stop sequences, and streaming. Check the API docs for the exact supported surface before treating it as a complete drop-in replacement.

How does billing work?

It's pay-as-you-go on prepaid credits — $0.50 per 1M input tokens and $2.00 per 1M output tokens, with no monthly subscription. Both prompt and completion tokens count. The gateway uses provider-reported usage when available and a documented estimate when it is not.

B/GPT

THE GATEWAY IS OPEN.

Break the black box.

Test the product path now. Connect production inference when its identity is ready to publish.

Create an account