KoboldAI Lite custom API setup (OpenAI Compatible API)

• A browser. Lite keeps its stories in the browser's local storage; the Save / Load menu exports them.
• A RelaySwitch key from the console (email code sign-in). Keys start with sk-relay-.
• A balance. Top-ups start at $10, by card or crypto, with no subscription. Before the first top-up the console chat runs 12 free turns a day.

Click AI in Lite's top bar. The Select your AI provider popup opens; set these:

1. Provider: OpenAI Compatible API (under the custom endpoints, not the KoboldAI Remote API).
2. OpenAI API URL: https://api.relayswitch.io/v1. Keep Add Ver. Num as it is; the URL already carries the /v1.
3. API Key (Required): your sk-relay- key.
4. Click Fetch. Lite loads the model list your key can call from /v1/models; pick one. Or choose Custom and type the id exactly as the Models page prints it.
5. Tick Chat-Completions API. It sends each turn as a chat request, which is what every model here answers; unticked, Lite uses the text-completions route and gets 404.
6. Leave Use CORS Proxy unticked. The API accepts calls from any origin, so the key travels from your browser to the API and nowhere else.
7. Click Ok. The model name appears in the top bar, and the first message is the test.

The Favorites row in the same popup saves the URL and key for next time.

• Usage Mode: Story, Adventure, Chat or Instruct. Chat and Instruct map cleanly onto chat completions.
• Max Output: the longest reply per turn. 300 is a paragraph; a model that reasons before it writes needs 1,000 or more, or the reply is cut inside the reasoning step and arrives empty.
• Context Size: how much of the story Lite sends each turn. It is the number that sets the input side of the bill; keep it at or under the model's window.

Uncensored and roleplay builds first; the flagships for longer memory. Prices are per million tokens, and the turn columns price one 2K-in / 300-out turn: 2,000 tokens of story and memory in, 300 tokens out.

The uncensored models page lists every build the catalog tags.

Per message at three Context Size values, with a 300-token reply:

Lite's Context panel holds Memory and World Info, which ride along on every turn too. The console's Usage page lists every request with its tokens and charge.

• 401: the key is wrong or revoked. Paste it again in API Key (Required).
• 402: the balance is empty, or the key reached its budget. Top up or raise it in the console.
• 404 route_not_found: Chat-Completions API is unticked, or the URL lost its /v1. model_not_found: a typed id that is not in the catalog; use Fetch.
• A browser CORS error: Use CORS Proxy was ticked and the proxy is down. Untick it; this API needs no proxy.
• 429: the key hit its per-minute limit. Wait a few seconds. 503: no host for that model right now; pick another.

Labels from the provider popup at lite.koboldai.net and its source, LostRuins/lite.koboldai.net, read 2026-11-03. KoboldCpp users: the same popup appears when KoboldCpp is launched with --nomodel, so a local install can use this key the same way.