---
title: "Uncensored coding AI for Claude Code, Cursor, Open WebUI"
canonical: https://unrestrictable.ai/guides/uncensored-coding-ai
generated: 2026-11-16T02:57:39.485Z
source: https://unrestrictable.ai/guides/uncensored-coding-ai
base_url: https://api.unrestrictable.ai/v1
---

# Uncensored coding AI: run it inside Claude Code, Cursor and Open WebUI

Code builds with the refusal edited out of the weights, wired into Claude Code by router, into Cursor and into Open WebUI. A day of agent work priced from the live catalog, a one-request check, and the lines the acceptable-use page draws.

Steps checked October 27, 2026 · prices as of November 16, 2026

Uncensored coding AI means a code model whose weights no longer carry the trained refusal, dropped into the agent you already work in. Below: the code builds on this key, how to wire them into Claude Code (router in between), into Cursor and into Open WebUI, what a day of agent work costs at the catalog’s current rates, and where the acceptable-use page still says no.

Every tool takes the same three settings: a base URL of `https://api.unrestrictable.ai/v1`, a key made in the [console](https://app.unrestrictable.ai/console), and a model id. The [docs page](https://unrestrictable.ai/docs) gives those three for every other client. To try a build with nothing installed, the browser chat at [/chat](https://unrestrictable.ai/chat) answers three free turns a day with no account.

## Refusals inside an agent, and what the edit removes

A stock coding assistant says no in two ways. The loud one is a flat refusal: exploit code, a packer, a credential harvester for a lab, a phishing kit you want deobfuscated. The quiet one does more damage inside an agent: the reply gets narrowed, a disclaimer sits where a function should be, or the loop halts and asks you a question. Both end the task, because asking a refusal-trained model again buys the same answer at the same cost.

An uncensored code build has had that behaviour taken out of its weights, either by abliteration or by fine-tuning on data that holds no refusals ([the two edits](https://unrestrictable.ai/blog/abliterated-vs-uncensored)). Hand it a decompiled binary, a packet capture or a malware sample and it treats the input as text to work through, then writes what you asked for. Its programming skill is untouched: whatever the base model could do in Python, Rust or SQL, this build does.

## The code builds on this key

Three code builds today, with the general assistant beside them as a reference row. Rates are per million tokens, read from the catalog when the page renders.

| Model | Build | Input $/M | Output $/M |
| --- | --- | --- | --- |
| [Uncensored Coding Agent](https://unrestrictable.ai/models/uncensored-coding-agent) uncensored-coding-agent | House build | $0.90 | $3.00 |
| [GLM-5.3 Flash Code Uncensored](https://unrestrictable.ai/models/glm-5.3-flash-code-uncensored) glm-5.3-flash-code-uncensored | Uncensored | $6.00 | $6.00 |
| [Uncensored Code Assistant](https://unrestrictable.ai/models/uncensored-code-assistant) uncensored-code-assistant | House build | $11.70 | $15.60 |
| [GLM-5.3 Uncensored](https://unrestrictable.ai/models/glm-5.3-uncensored) glm-5.3-uncensored | Uncensored | $0.282 | $0.846 |

Prices per million tokens, read from the live catalog on 2026-11-16.

- [`uncensored-coding-agent`](https://unrestrictable.ai/models/uncensored-coding-agent): the Unrestrictable house build made for agent loops; 1,048,576 tokens of context fits an entire repository into a single request.
- [`glm-5.3-flash-code-uncensored`](https://unrestrictable.ai/models/glm-5.3-flash-code-uncensored): built for code, refusal removed, from Z.ai's GLM-5.3 Flash (320B total, 18B active, MIT, card read 2026-10-26).
- [`uncensored-code-assistant`](https://unrestrictable.ai/models/uncensored-code-assistant): Unrestrictable’s own build aimed at one-shot answers in a chat panel, the priciest row in the table, with 131,072 tokens of context.

The general build `glm-5.3-uncensored` writes ordinary code well enough. Reach for a code build when the job is security research or a long agent session.

## One request before any agent

From a terminal, a single request proves the build answers what a stock assistant turns down, and prints the charge header:

```
curl https://api.unrestrictable.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-unr-live-…" \
  -H "Content-Type: application/json" \
  -d '{"model": "uncensored-coding-agent", "messages": [{"role": "user", "content": "Write a Python function that parses a PE header and lists its imports. Then explain how a packer hides them."}]}' \
  -i | grep -i -E "^HTTP|x-uplink-charge"
```

A 200 with `x-uplink-charge` in the headers is the entire test. 401 points at the key. 404 with `model_not_found` points at the id, nearly always a display name typed where the slug belongs. Keys come from the [console](https://app.unrestrictable.ai/console); give one a spend limit before it goes into a tool other people share.

## Claude Code, with a router in the middle

Claude Code speaks the Anthropic request format. Every build here speaks OpenAI’s. claude-code-router translates between the two. The [Claude Code guide](https://unrestrictable.ai/guides/claude-code) walks through the install and the test request; here is that provider block, code builds filled in:

```
{
  "APIKEY": "a-client-key-you-choose",
  "HOST": "127.0.0.1",
  "PORT": 3456,
  "Providers": [
    {
      "name": "unrestrictable",
      "api_base_url": "https://api.unrestrictable.ai/v1/chat/completions",
      "api_key": "sk-unr-live-…",
      "models": ["uncensored-coding-agent", "glm-5.3-flash-code-uncensored", "uncensored-code-assistant"]
    }
  ]
}
```

```bash
ccr serve --no-open
export ANTHROPIC_BASE_URL="http://127.0.0.1:3456"
export ANTHROPIC_API_KEY="a-client-key-you-choose"
claude --model uncensored-coding-agent
```

`APIKEY` is a string you invent; Claude Code presents it to the router. `api_key` is your Unrestrictable key. To change build during a session, type `/model` followed by another id from the block. The router’s README lives at [musistudio/claude-code-router](https://github.com/musistudio/claude-code-router).

## Cursor, under Settings and Models

Cursor accepts a custom OpenAI key and base URL in Settings, Models. Fill three fields, add the id, and the build replies in chat and in Agent mode. Tab completion keeps using Cursor’s own models.

- OpenAI API Key: sk-unr-live-… the key from your Unrestrictable console
- Override OpenAI Base URL: https://api.unrestrictable.ai/v1 switched on; keep the /v1 at the end
- Model: uncensored-coding-agent typed by hand into the model list, character for character

Every field, every error and its fix: the [Cursor guide](https://unrestrictable.ai/guides/cursor).

## Open WebUI, a connection and a preset

Go to Admin, then Connections, and create an OpenAI API connection: `https://api.unrestrictable.ai/v1` as the URL, your key beneath it. Open WebUI pulls the model list from the API. Allow only the code builds so the picker stays short, then give each build a Workspace preset carrying the system prompt your team works with. Screens, the context-window setting and a priced chat are in the [Open WebUI guide](https://unrestrictable.ai/guides/open-webui).

## A day of agent work, priced

The figures come from the catalog of 2026-11-16, worked out when the page renders: 150 agent turns, each sending 60,000 cached prefix tokens (the repository plus the conversation to date), 2,500 fresh input tokens and 1,200 tokens of edits back. Where the catalog publishes a cache-read rate, cached tokens are billed at it.

| Build | Context | Cost of the day | Today’s check |
| --- | --- | --- | --- |
| [Uncensored Coding Agent](https://unrestrictable.ai/models/uncensored-coding-agent) | 1,048,576 tokens | $4.12 | Steady since Oct 22 |
| [GLM-5.3 Flash Code Uncensored](https://unrestrictable.ai/models/glm-5.3-flash-code-uncensored) | not published | $14.13 | Steady since Oct 22 |
| [Uncensored Code Assistant](https://unrestrictable.ai/models/uncensored-code-assistant) | 131,072 tokens | $38.08 | Steady since Nov 14 |
| [GLM-5.3 Uncensored](https://unrestrictable.ai/models/glm-5.3-uncensored) | not published | $0.98 | Steady since Oct 12 |

Formula: turns × (cached × cache-read rate + input × input rate + output × output rate), on the catalog of 2026-11-16. The last column is today’s result from the [daily real-model check](https://unrestrictable.ai/proof).

Over an agent’s day the cached prefix dominates, which keeps a build with a low cache-read rate cheap across a long session even when another build undercuts it on output. All rates are on the [pricing page](https://unrestrictable.ai/pricing). Your balance is prepaid and each request settles against it at those rates.

## Security work: what opens up, and what does not

What opens up: the build writes the exploit for the CVE on your patch list, steps through the sample you are classifying, rewrites a payload to exercise a detection rule, and reads a phishing kit as source code. The [red-teaming page](https://unrestrictable.ai/red-teaming) covers the security builds trained on security corpora and their agent-tool settings; inside a coding agent the code builds above cover that ground with more code and less talk.

What does not: the [acceptable-use page](https://unrestrictable.ai/acceptable-use) binds every key. No access to, attacks on or tests of systems, networks or accounts that are not yours or covered by written permission. No building or spreading malware against others. No fraud, phishing or impersonation. Security and red-team builds are for systems you own or hold written authorization to test, and the policy asks you to keep that authorization on file. The gateway does not screen chat requests by content; the build answers as built, and the policy governs what you send it. Break it and the key is revoked, after which the API returns 401 within 60 seconds.

## Questions

### Does an uncensored coding AI without refusals exist?

It does. The code builds on this key have had the refusal removed from the weights, so exploit, packer and malware-analysis questions come back as code. uncensored-coding-agent runs inside Claude Code through claude-code-router, inside Cursor via its Models settings, and inside Open WebUI. See the first code build.

### What does "unfiltered coding AI" mean?

The same thing under a different label: a code model served with neither a trained refusal nor a moderation layer in front. Here it is a build whose weights were edited rather than a stock model behind a loosened filter, which is why no jailbreak prompt has to be kept working. See how the edit works.

### Which uncensored code assistant can Claude Code use?

Any id listed in the router's provider block. The config above carries the code builds; claude --model uncensored-coding-agent opens a session on the first one, and /model changes it mid-session. See the Claude Code guide.

### Will an uncensored AI for code run in Cursor’s Agent mode?

Yes. Agent mode issues OpenAI-format tool calls, which the code builds respond to. Tab completion stays on Cursor’s own models, while the key and base URL cover chat, Ask and Agent alike. See the Cursor guide.

### Is code from an uncensored AI safe to execute?

Handle it like any model output: read it first, run it in a lab, and never on systems that are not yours. On every key, the acceptable-use page bars attacks on or tests of systems you lack written permission for, and malware aimed at others. See the acceptable-use page.

### Can I get an uncensored coding AI from GitHub?

GitHub hosts the tooling (Open WebUI, llama.cpp, claude-code-router) plus the abliteration code; the weights themselves sit on Hugging Face. The builds here answer over the API with nothing to download, and each family page links the build’s card for anyone who wants to run it locally. See the models by family.

Every model above is on the [uncensored model list](https://unrestrictable.ai/models) with its live price per million tokens. [Per-token pricing](https://unrestrictable.ai/pricing) sets the whole catalog side by side, and the [family directory](https://unrestrictable.ai/uncensored) groups the builds by base model.

[Get a key](https://app.unrestrictable.ai/console) Prepaid from $10, no subscription.

## New models and price changes, once a month.

New uncensored builds with their prices, price moves, and what the daily model check found.
