Uncensored AI Chat Online With No Filter: Browser, App, or API
We build the local app first. But a lot of people do not have a 24 GB card, and the question that reaches our Discord every week is the same one: can I get this in a browser instead. Yes. This guide covers the three routes that actually work, with the model names, the prices, and the limits written out plainly. If you want the short version of what "no filter" means before you spend anything, read what no filter means here first, or the overview page at uncensored AI chat online.
Route 1: The Browser at lu-labs.ai
LU Labs Cloud is the hosted layer from the same team that ships Locally Uncensored. Nothing to install:
- Sign in. Create an account at lu-labs.ai or sign in to one you already have.
- Open Chat. The Studio holds chat, image and video in one window. Chat is the first tab.
- Pick a model. The picker lists the chat catalog with a cost per token next to each entry.
- Start typing. Credits come off per token as the answer streams. Switching models mid thread is fine.
Which model for what
Fifteen chat models is enough to be confusing, so here are the ones that matter for this use.
| Model | Use it for | Note |
|---|---|---|
| DeepSeek V3.2 | The cheap everyday default | 0.038 credits per output token, think toggle |
| Kimi K3 | Hard prompts, long instructions, agent work | Reads images, 1M token context, native tool calls |
| Euryale 70B | Long form character and roleplay | Community fine tune |
| Lunaris 8B | Fast roleplay on a small budget | Cheapest of the character builds |
| MythoMax 13B | Classic storytelling | Old but still requested by name |
| GLM 5.3 Flash | Vision input on a long context | Reasons on every turn, 1M token context |
Start on Lunaris 8B or DeepSeek V3.2, move to Kimi K3 when a prompt is genuinely hard, and switch to a community fine tune when you want a character voice rather than an assistant voice. One catalog note: the picker is the same on every plan and on a credit pack, so the big entries such as Qwen 3.8 Max and Qwen 3.8 A95B are in the list on a 5 EUR pack too. They just cost more per token, at 0.4951 and 0.6 credits per output token against 0.038 on DeepSeek V3.2.
Route 2: The Desktop App With the Cloud Switch
The desktop app is free, open source under AGPL-3.0, and runs on Windows and Linux. It also has a Cloud switch that signs into the same LU Labs account. Flip it and the hosted models appear in the same picker as your local ones, so a 7B running on your own GPU and Kimi K3 running on someone else's sit two rows apart in the same dropdown. That is the setup we use ourselves: local for anything sensitive, hosted for the models that will never fit in consumer VRAM.
There is no native Mac build. On a Mac you use the browser Studio from Route 1, which is the same product in a tab.
Route 3: An API Key for Your Own Client
If you already have a front end you like, skip both UIs. In your LU Labs account open Settings > Cloud API keys and create a key. It carries an lu_ prefix and is shown exactly once, so paste it somewhere safe before you close the dialog.
- Base URL:
https://lu-labs.ai/api/inference/v1 - Shape: OpenAI chat completions, so any client that speaks it works unchanged
- Model id example:
moonshotai/Kimi-K3
For the readers who came here for roleplay, this is the part you want. In SillyTavern, choose Chat Completion as the API, pick a custom OpenAI compatible source, paste the base URL and the key, and type the model id by hand. Two things trip people up: SillyTavern's Text Completion mode is the wrong one here, and the model id has to be typed exactly as written, capitals included. Aider, LibreChat and plain curl are all known to work against the same endpoint, so a two line smoke test is the fastest way to prove the key before you wire up a front end.
curl https://lu-labs.ai/api/inference/v1/chat/completions \
-H "Authorization: Bearer lu_YOUR_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"moonshotai/Kimi-K3","messages":[{"role":"user","content":"Say hello."}]}'If that returns a completion, every client on the list above will connect with the same three values.
What It Costs
No subscription is required. Any pack can be your first purchase and credits never expire.
| Option | Credits | Shape |
|---|---|---|
| 5 EUR pack | 165,000 | One off, no subscription |
| 10 EUR pack | 350,000 | One off |
| 25 EUR pack | 900,000 | One off |
| Hosted month | 900,000 | 19 EUR a month |
The number that makes this concrete: Euryale 70B costs 0.085 credits per output token, so the 5 EUR pack is roughly 2.3 million output tokens on it. That is a lot of conversation for the price of a coffee, and it does not expire if you use it in bursts. Current numbers live at https://lu-labs.ai/pricing.
The Limits, Written Straight
We would rather you read this before you pay than after.
- Hosted text models answer as they were trained. We add no moderation layer on top of them. That is a claim about our stack, not a promise about any given model's behaviour.
- Hosted image and video follow the Terms of Service, and the prompt is checked before it renders. The picture lane is not the text lane.
- Content involving minors is refused everywhere, always. No exception, no route around it, on any surface.
- The hardware is shared. At busy hours there is a short queue before your request starts.
- No training on your prompts, never sold. Accounts and billing sit in the EU.
- The catalog is curated. You cannot upload your own checkpoint to the hosted side.
The Fully Private Option
Hosted chat is a convenience, not the strongest privacy answer, because a hosted request leaves your machine by definition. If that matters more to you than hardware cost, run the free app locally with an abliterated model and nothing in the middle: no account, no bill, no request that leaves the box. Our walkthrough is how to run uncensored AI locally, and the model picks with direct download links are in the abliterated models guide. If you have the GPU, take that route. If you do not, Route 1 gets you a conversation in about two minutes.