KoboldAI Lite custom API: uncensored models, no GPU

What KoboldAI Lite is

KoboldAI Lite is a web page for writing stories, playing text adventures and chatting with characters. It runs in the browser at lite.koboldai.net and is also the interface bundled with KoboldCpp, the local runner. Its source is public on GitHub at LostRuins/lite.koboldai.net.

Lite has no model of its own. It talks to a backend: a KoboldCpp server on your machine, the AI Horde, or a hosted API. The OpenAI Compatible API provider is the hosted route. With it, the model runs on the API, so you need no GPU, no model download and no KoboldCpp. The labels in this guide come from Lite's source as published on 24 October 2026.

The KoboldAI Lite custom API settings

Everything sits in one popup. These are the values for this API.

Provider

OpenAI Compatible API

In the Select your AI provider list

OpenAI API URL

https://api.uncensoredllm.ai/v1

Keep Add Ver. Num ticked

API Key (Required)

sk-ullm-live-…

Created in the console, shown once

Model

cydonia-24b-v4.3

From Fetch, or typed with Custom

Chat-Completions API

ticked

Use CORS Proxy

unticked

Connect KoboldAI Lite to the uncensored API

1. Create a key in the console. It starts with sk-ullm- and is shown once. The balance is prepaid from $10; a key with nothing on it is accepted but the first turn answers 402.
2. Open lite.koboldai.net and click AI in the top navigation. The Select your AI provider popup opens.
3. In the provider dropdown, choose OpenAI Compatible API.
4. Paste https://api.uncensoredllm.ai/v1 into the URL field (its placeholder reads OpenAI API URL). Keep Add Ver. Num ticked. Lite's own note on that box: "OpenAI Compatible APIs end with /v1. If they do not, uncheck this box." This one does.
5. Paste your key into the field marked API Key (Required).
6. Click Fetch to load the model list your key can reach and pick an id, for example cydonia-24b-v4.3. Or click Custom and type the id exactly as the model list shows it.
7. Tick Chat-Completions API. This sends each turn as chat messages to /chat/completions, the endpoint every model here answers on. Leave it unticked and Lite uses the older text endpoint, which this guide does not cover.
8. Leave Use CORS Proxy unticked. Lite's popup says it plainly: "API keys are used with OpenAI API directly, unless CORS proxy is used." Unticked, the key travels only from your browser to the API.
9. Click Ok. Lite connects and the model is ready.

The Favorites row in the same popup saves the URL and key pair in your browser, so you do not paste the key again next time. Skip it on a shared computer.

Context Size, Max Output and Usage Mode

Three settings in Lite's Settings panel decide how a session reads and what it costs.

• Usage Mode: Story, Adventure, Chat or Instruct. Pick the one that matches the session; the connection stays the same.
• Max Output: the longest reply per turn. We start at 512 tokens for adventure and chat.
• Context Size: how much of the story Lite sends on each turn. Keep it under the model's window minus Max Output.

Lite's popup also notes that only Temperature, Top-P and Repetition Penalty samplers are used with this provider. Ticking Non-Standard Fields sends the rest (Min-P, Top-K and others); Lite warns that many endpoints reject them, so leave it off unless you need one.

Lite's Context panel holds Memory and World Info. Memory is text sent on every turn; World Info entries are sent when their keys appear in the story. Both count against Context Size and both are billed as input on each turn they ride along. A 500-token Memory costs 500 tokens on every one of the 150 turns in the example below, so keep it to the facts the story cannot lose.

Temperature is the sampler to adjust first. Lower it when a build drifts off the plot; raise it when replies repeat themselves. Repetition Penalty a little above 1 helps a small model that loops on a phrase over a long adventure.

Which uncensored build to pick for Kobold Lite

All four builds below carry the roleplay tag in the catalog and take the OpenAI wire format. Prices and windows are read from the live catalog when the page renders.

Context windows read from the live catalog on 2026-10-27. Context Size is the window, capped at 16,384, minus a Max Output of 512.

mythomax-l2-13b has a 4,096 tokens window, so Context Size has to stay under it: a long story falls out of memory quickly there. The 24B and Gemma 4 builds take far more history. The Gemma tune is one of several; the rest are on the uncensored Gemma page.

What an adventure session costs

The worked example: 150 turns in Adventure mode, each reply about 200 tokens. The short column sends about 2,000 tokens of story per turn; the long column sends 8,000, a longer memory and world info.

Computed at render from the catalog prices on 2026-10-27: turns × prompt tokens × input price, plus turns × 200 × output price. "Over the window" means that fill does not fit the model.

Input dominates: 150 × 2,000 is 300,000 input tokens against 30,000 output tokens. Context Size is the setting that moves the bill. Every per-token rate is on the pricing page.

Check the key and the model id first

This request lists the model ids your key can reach. It is the same call Lite's Fetch button makes, it costs nothing, and it tells you whether the key works before you open Lite. Swap in your key and run it from a terminal.

A 200 with a list of ids means the key is good. A 401 means the key, and the message says which way:

Both bodies as the API returned them on 24 October 2026.

Troubleshooting Kobold Lite errors

401: the key

"No API key was sent." means the key field is empty. "The API key is not valid." means it is wrong, cut short or revoked. Create a fresh key in the console and paste it whole.

404 model_not_found: the model id

The id is not in the catalog: a typo in Custom, or a build that has since left. Click Fetch again and pick from the list.

404 on every turn: Chat-Completions API is off

If Fetch works but every turn fails, check that Chat-Completions API is ticked. Without it Lite calls the text-completion endpoint instead of /chat/completions.

402: the balance

The prepaid balance is at zero. Top up in the console; a request refused with 402 is not charged.

Browser errors (CORS)

Lite calls the API from your browser. The API allows calls from any origin, including the Authorization header (checked on 24 October 2026), so the CORS proxy is not needed and should stay off.

The daily real-model check sends a fixed prompt to every model each day, so the model behind an id is the one on its label. Base URL, key and model id for other clients are on the docs page.

Questions