# uncensored-ai-api (UncensoredLabs/uncensored-ai-api)

> A hosted API that gives you uncensored and abliterated DeepSeek and GLM text models through one OpenAI-compatible endpoint, plus a browser Chat. This repository only holds guides, configuration examples and API samples.

- GitHub: https://github.com/UncensoredLabs/uncensored-ai-api
- GitHub.Locker page: https://github.locker/UncensoredLabs/uncensored-ai-api
- Stars: 682
- Language: JavaScript
- License: MIT
- Last push: 2026-11-01

## Questions and answers

Written by GitHub.Locker from the README and the repository; check the README below for details.

**Is it something I install, or a service I sign up for?**

It's a hosted service: you sign up, create an application API key, and point your existing OpenAI-compatible client at it. Nothing here is self-hosted, because the platform code itself is not in this repo.

**How do I make my first request?**

Create an account and an application API key in the dashboard, keep your existing client, change the base URL and choose a model ID. Don't use the 16-digit sign-in key in API requests.

**What do I need on my side?**

Mostly an OpenAI-compatible SDK: the examples are in Python and TypeScript, and the guides also cover configuring coding clients such as Cline, Continue and OpenCode. The quickstart has separate Linux/macOS and PowerShell setup steps.

**Is there a catch?**

'Abliterated' models have modified weights, and they are still not guaranteed to answer everything. Repeated trial sign-ups may need identity verification, and a low balance can block requests because each request reserves credit before generation.

**What do people actually build with it?**

The README points to security research (vulnerability chains, regression tests), reverse engineering of decompiled code, coding-assistant integrations, creative writing on hard themes, and trust and safety evaluation.

**Does it support streaming responses?**

Yes. The streaming guides cover incremental text, usage events and stream failures, with runnable examples in both Python and TypeScript.

**What happens when a request fails or my balance runs low?**

The errors and billing guide covers HTTP errors, rate limits, credit reservations and retry decisions, so check your dashboard balance before starting a large job.

**Can I connect a coding assistant to it?**

Yes. The guides include configuration for Cline, Continue and OpenCode, so the coding client can reach the models through the same OpenAI-compatible API.


---

## README (checked against GitHub 2026-11-05)

Source: https://github.com/UncensoredLabs/uncensored-ai-api/blob/main/README.md

![UncensoredLabs — uncensored models, one API](assets/banner.svg)

# Uncensored AI API — UncensoredLabs

Access uncensored and abliterated **DeepSeek and GLM text models** through one OpenAI-compatible API, or use them directly in browser Chat. Keep your client, change the base URL, and choose a model.

[Get started](https://uncensoredlabs.dev/signup?utm_source=github&utm_medium=repository&utm_campaign=uncensored-ai-api) · [Models and pricing](https://uncensoredlabs.dev/models?utm_source=github&utm_medium=repository&utm_campaign=uncensored-ai-api) · [Browser Chat](https://uncensoredlabs.dev/dashboard/chat?utm_source=github&utm_medium=repository&utm_campaign=uncensored-ai-api) · [Documentation](https://uncensoredlabs.dev/docs?utm_source=github&utm_medium=repository&utm_campaign=uncensored-ai-api)

Built in Switzerland. This is the official public repository for **guides, configuration examples and API samples**. The hosted platform's implementation is not included.

## Start with a working request

Create an account and an **application API key** in the dashboard. A 16-digit account sign-in key is a different credential; do not use it in API requests.

```bash
export UNCENSOREDLABS_API_KEY="YOUR_APPLICATION_API_KEY"
curl --fail-with-body https://uncensoredlabs.dev/v1/chat/completions \
  -H "Authorization: Bearer $UNCENSOREDLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4.1-flash-1017-abliterated",
    "messages": [{"role": "user", "content": "Review this security design: customer comments are rendered as HTML in an admin console. Explain the attack path and how to verify the fix."}],
    "max_tokens": 1024
  }'
```

Eligible new accounts receive **$0.10 in trial credits**; repeat-trial signals can require identity verification. Prepaid top-ups start at **$3**. Check the dashboard's available balance: requests reserve credit before generation, and a small balance may not cover every model's reservation.

## Models

Snapshot checked **November 1, 2026** against the [public API catalog](https://uncensoredlabs.dev/v1/models). USD per **one million tokens**, before any active plan discount. Context is 1,000,000 tokens for every model below; input and output must fit the provider's context accounting.

| Exact API model ID | Input | Cached input | Output | Maximum output |
| --- | ---: | ---: | ---: | ---: |
| `deepseek-v4-pro-uncensored` | $0.50 | $0.10 | $0.50 | 65,536 |
| `deepseek-v4.1-flash-1017-abliterated` | $0.30 | $0.12 | $0.30 | 65,536 |
| `deepseek-v4.1-flash-abliterated` | $0.30 | $0.10 | $0.30 | 65,536 |
| `deepseek-v4.1-flash-abliterated-fast` | $1.40 | $0.32 | $1.40 | 65,536 |
| `glm-5.3-abliterated` | $1.40 | $0.40 | $1.40 | 130,000 |
| `glm-5.3-uncensored` | $0.35 | $0.10 | $0.35 | 65,536 |

All six catalog entries support text, streaming and function tools. **Catalog inclusion does not mean a model is currently healthy.** See [model details and performance](docs/models.md) for live links and measurement limitations. Rates and availability can change; the live catalog takes precedence.

## What to build

- **Security research:** reason through vulnerability chains, review application trust boundaries, and turn findings into regression tests.
- **Reverse engineering:** explain decompiled routines, recover file-format structures, and compare a vulnerable function with its patch.
- **Developer tools:** connect a coding assistant, process tool calls, or integrate streamed text into your own interface.
- **Creative work:** explore difficult themes and unconventional characters with models designed for fewer refusals.
- **Trust and safety:** create labeled evaluation cases and analyze moderation failure modes.

“Abliterated” describes models whose weights have been modified to reduce refusal behavior. It does not promise a correct answer or no refusals on every prompt. UncensoredLabs is an independent service, not affiliated with OpenAI, DeepSeek or Zhipu AI.

## Guides and examples

| Guide | What you get |
| --- | --- |
| [Quickstart](docs/quickstart.md) | Linux/macOS and PowerShell environment setup; first request |
| [Python](examples/chat.py) | OpenAI SDK request with automatic retries disabled |
| [TypeScript](examples/chat.ts) | OpenAI SDK request with typed error handling |
| [Streaming](docs/streaming.md) | Incremental text, usage events and stream failures |
| [Streaming Python](examples/stream.py) | Runnable streamed response |
| [Streaming TypeScript](examples/stream.ts) | Runnable streamed response |
| [Coding clients](docs/clients.md) | Cline, Continue and OpenCode configuration |
| [Model catalog](docs/models.md) | Limits, pricing, live performance and model page links |
| [Security research workflows](docs/security-research.md) | Concrete prompts for analysis and evaluation |
| [Errors and billing](docs/errors-and-billing.md) | HTTP errors, rate limits, reservations and retry decisions |
| [Compatibility and privacy](docs/compatibility.md) | Supported protocol and content handling |

## Stay in touch

[X / @uncensoredlabs0](https://x.com/uncensoredlabs0) · [Telegram bot](https://t.me/UncensoredLabsDevBot) · [Website](https://uncensoredlabs.dev/?utm_source=github&utm_medium=repository&utm_campaign=uncensored-ai-api)

Use [issues](https://github.com/UncensoredLabs/uncensored-ai-api/issues) for reproducible documentation problems and example improvements. Keep account, payment and vulnerability details out of public issues; see [support and security reporting](SECURITY.md).

## License

The original documentation and example code in this repository are [MIT licensed](LICENSE). That license does not cover hosted platform source, model weights, provider services or third-party trademarks.