Use KoboldAI Lite with an uncensored OpenAI-compatible API
KoboldAI Lite is a single-page writing and chat client, hosted at lite.koboldai.net and bundled with KoboldCpp. It runs entirely in your browser, so on its own it cannot reach the Wild West API, but its built-in Use CORS Proxy option gets around that.
Set it up
- Sign in at wildwestapi.com/signup/ with the six-digit code sent to your email, open the dashboard and create an API key. It starts with sk-ww- and is shown in full once, so copy it somewhere safe.
- Open KoboldAI Lite and click AI to open the connection dialog.
- Choose OpenAI Compatible API from the provider list.
- Enter https://wildwestapi.com/v1 as the OpenAI API URL and paste your key into the API Key field.
- Tick Use CORS Proxy. Leave Add Ver. Num ticked (it only adds /v1 when it is missing).
- Choose [Custom] in the model list and enter outlaw-1, or use Fetch List once the proxy box is ticked.
- Keep Chat-Completions API ticked and click Connect.
Does it work with the Wild West API?
Only with the CORS proxy. Lite has no server of its own: the page in your browser makes every request. Browsers send a preflight check before a cross-site request that carries an API key, and the Wild West API does not answer that check today, so a direct connection fails with a network error.
Lite's Use CORS Proxy box routes the request through a small relay run by the KoboldAI Lite developer, which returns the headers the browser wants. Lite's own help text describes it as a way to "bypass CORS restrictions" for public internet APIs. With it ticked, the setup works. Your key and prompts then pass through that relay, so decide whether you are comfortable with that. If not, use SillyTavern, which calls the API from its own server.
The settings that matter for long roleplay
- Context Size in Settings caps how much story is sent. The slider tops out at 16384, but you can type a larger number in the box. These models take 512K tokens or more; see context window.
- Max Output is the reply length, the same as max tokens.
- Token streaming shows text as it arrives. If streaming misbehaves through the proxy, Lite has a Disable CORS proxy streaming option.
- Lite's help notes that only Temperature, Top-P and Repetition Penalty are sent to OpenAI-style APIs, so other sampler sliders do nothing here.
Troubleshooting KoboldAI Lite
- 401: the key is missing, mistyped or disabled. Paste it again with no trailing space and check it starts with
sk-ww-. A key from another provider will never work here. - 402: your balance is too low for the request. Top up in the dashboard; see pricing. A huge context on every turn burns credit faster than you might expect.
- 404 or 400 with an unknown model: the model id is wrong. Use an exact id from the model list, such as
outlaw-1, with no provider prefix and no spaces. - 429: you are rate limited. Wait a few seconds and send again, and turn off any auto-retry loop that fires instantly.
- 5xx: trouble upstream. Retry; if it keeps happening, switch to another model id for a while.
- Network error, "Failed to fetch" or a CORS message: the Use CORS Proxy box is not ticked. Tick it and connect again. Running Lite from KoboldCpp does not change this, since the page still makes the request.
- Model list is empty: Fetch List also needs the proxy. Pick [Custom] and type the id instead.
FAQ
Why does KoboldAI Lite need a proxy when SillyTavern does not?
SillyTavern runs a small server on your machine that makes the API call. Lite is only a web page, so the browser makes the call and enforces CORS.
Does the bundled Lite inside KoboldCpp avoid the problem?
No. KoboldCpp serves the page, but the page still calls the remote API from your browser, so the same CORS rule applies.
Will this keep working if the Wild West API adds CORS support?
Yes. If the API starts answering preflight, you could untick Use CORS Proxy and connect directly.