Venus AI API setup: connecting your own key on chub.ai

Quick answer: the three values Venus needs

Venus treats a custom API connection as a core feature, not an afterthought, since chub.ai does not want to run inference for every character card itself. Open a chat, go into the connection or generation settings, switch the source to Custom or OpenAI-compatible, and fill in three values: an endpoint URL that points at a chat completions API, an API key for that endpoint, and the exact model ID you want to run. Leave the model ID as a display name instead of the real ID and Venus either queues whatever it defaults to or returns a plain error that does not say which field is wrong. The endpoint format is the detail that trips people moving between roleplay tools. Some Venus builds want only the base URL, ending at /v1, and append the rest of the path themselves. Others want the full path ending in /chat/completions, the way a proxy field works. If one format returns a 404, the other is almost always the fix, so keep both forms written down the first time a config gets saved.

How the connection works, and which models people actually pick

Once connected, Venus behaves like every other bring-your-own-key roleplay frontend: it resends the character card, the persona, and a rolling slice of chat history with every message, then appends the new line at the end. None of that is optional and none of it is cached by Venus itself, so the token count of a single message is really the token count of everything the model needs to remember about the scene so far. That mechanic is why the model choice matters more than almost any other setting. A model priced for high-volume chat can carry a long-running scene for pennies. A flagship model built for careful reasoning bills the same premium rate whether it is debugging code or answering one line of dialogue, because input tokens get charged at the same price regardless of the task.

• Budget, high-volume chat: DeepSeek V4 Flash sits at the bottom of the price table and holds up fine for casual back-and-forth.
• Longer or more detailed scenes: DeepSeek V4 Pro and Kimi K2.6 keep more history coherent before quality starts to slip.
• Distinct voice, mature fictional themes: Grok 4.5 comes up often here, since xAI is one of the few major providers whose stated policy is generally read as leaving room for that in fiction, though wording shifts over time and is worth rechecking.
• Prose quality: Claude Sonnet 4.6 is frequently rated the best stylist on this list, though Anthropic's terms rule out explicit content outright, which keeps it to SFW scenes only.

Worked example: what a month of chatting actually costs

Venus resends the card plus recent history on every message, so input tokens drive the bill more than output does. A setup with a detailed character card typically runs somewhere around 5,000 input tokens and 300 output tokens per message once a scene has some history behind it. The table below prices that out across 1,000 messages, a heavy month for most people; a lighter month scales down roughly in proportion. These are APIsRouter catalog rates, shown here as a reference point rather than the only source, from a pay-as-you-go gateway with no subscription to keep running. Whatever endpoint a config points at, the shape of the math stays the same: a budget model turns a month of chatting into pocket change, a flagship model turns it into a real line item.

Why the bill, or the errors, catch people off guard

These patterns account for most first-week support questions from card-chat users switching to their own key.

• Input, not output, drives the cost. The full character card and a slice of history get resent on every message, so a short reply can still be an expensive request.
• A silent 401 is almost always whitespace. A trailing space or line break left over from copy-paste is the single most common cause, followed by a key that was quietly revoked.
• Model IDs change under saved configs. Providers rename or retire IDs, and a config built around an old one starts failing without any change on the Venus side.
• Long scenes eventually push the oldest history out of the window. That reads as the character forgetting earlier events, not as a billing problem, but it comes from the same root cause: everything in context gets resent, and the window has a ceiling.
• Flagship and budget models can differ by more than 10x on the same message. That gap is invisible until someone compares the actual per-1,000-message math instead of a per-token headline number.

Venus vs SillyTavern vs JanitorAI: how the setup compares

All three tools solve the same problem, connecting a character-card chat to an outside model, but the details of the connection differ enough to cause confusion when someone switches between them.

Getting a setup that stays reliable and affordable

A few habits keep the setup stable long after the first session.

• Save one config per model, not one config that gets edited constantly. A dedicated slot per model ID means the only thing that changes when switching is which saved config is active.
• Test outside the app first. If a request fails inside Venus, the error surface rarely says which of the three fields is wrong; a direct request to the endpoint does.
• Re-paste keys with no surrounding whitespace, and re-save after any app update, since fields have been known to clear silently after one.
• Keep any custom or system prompt short. Chub-style cards already carry persona detail, and every extra sentence gets resent and billed on every message for the rest of the chat.
• Match the model to the scene: route casual, high-volume chatting to a budget model and save flagship pricing for scenes where prose quality is the point. Pointing the same connection at a lower-cost OpenAI-compatible endpoint is one way to bring the per-token rate down without changing anything else about how Venus is configured.

Config example: connect Venus and verify with curl

Before saving anything inside Venus, send one request straight to the endpoint with the same three values the app will use: the URL, the key, and the model ID. If this comes back with a completion, any failure afterward inside Venus is a UI or config problem, not a credentials problem.

curl https://api.apisrouter.com/v1/chat/completions \
  -H "Authorization: Bearer sk-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-v4-flash",
    "messages": [{"role": "user", "content": "Say ready."}],
    "max_tokens": 20
  }'

FAQ