Wyvern Chat custom API setup: one provider, every model.

Quick answer: register one custom provider.

In WyvernChat's settings, open the provider or connection section, and instead of picking a named service, create a custom provider. The values are the standard three of every OpenAI-compatible frontend: base URL https://api.apisrouter.com/v1, your sk-... key, and the exact model id you want the chat to run, such as deepseek-v4-flash or claude-sonnet-4-6. Field labels shift as the app evolves, so treat the names loosely and the values exactly. WyvernChat's documentation frames this capability directly: if your provider is not in the built-in list, you create a new provider and use the models it offers. That is the mechanism this whole page rides on, and it means model choice inside Wyvern stops being limited to the services the developers pre-wired.

Base URL:  https://api.apisrouter.com/v1
API key:   sk-...              (from APIsRouter)
Model:     deepseek-v4-flash   (typed exactly; any catalog id works)

What WyvernChat is, and why it wants your key.

WyvernChat (app.wyvern.chat) is a web roleplay platform in the SillyTavern lineage but hosted: character cards, lorebooks, group chats, and a public character library, with none of the self-hosting. Reviews and community write-ups consistently describe two things about it: the core platform is free without feature paywalls, and it deliberately supplies no models, so you bring a key from a provider and pay that provider directly. That BYOK design is the honest version of the deal most character platforms obscure. There is no bundled quota to run out of and no premium model tier to upsell; your costs are whatever your provider meters, visible in that provider's logs rather than hidden in a subscription. The flip side is that the provider decision is entirely yours, which is exactly the decision a multi-family gateway simplifies: one custom provider entry, one key, and the model question becomes a per-chat dropdown rather than a new account per vendor. Wyvern supports per-connection configuration beyond the basics, sampler presets and instruct templates among them, so a tuned setup for one model family can sit saved next to another. Where the docs do not spell out a detail (exact labels, or whether an Anthropic-native dialect is exposed alongside the OpenAI shape), assume the OpenAI-compatible custom provider is the dependable route; it is the one the platform's own documentation commits to.

Setup walkthrough.

Debugging follows the universal frontend pattern. A silent failure or immediate error with correct-looking values is usually one of three things: a URL pasted with the completions path attached when the app wants a base URL, a key with stray whitespace, or a browser-side CORS block, which you can spot in the developer console. Since Wyvern is a hosted web app, the request path runs from the app to your endpoint; proving the same three values work in a curl command splits endpoint problems from app problems in under a minute.

• Create or sign in to your WyvernChat account at app.wyvern.chat.
• Open settings and find the provider / API connection section.
• Choose to add a custom provider rather than a named preset.
• Enter the base URL https://api.apisrouter.com/v1 and paste your key.
• Set the model id exactly as the catalog lists it; if the app fetches a model list from the endpoint, pick from it, otherwise type the id.
• Save, open any character, and send one short test message to confirm the round trip.

Character cards: what V2 import means in practice.

Wyvern's wiki documents character cards as portable artifacts: cards download in the V2 card format embedded in PNG files, and importing an image that carries card metadata (a SillyTavern export, for example) prompts you to pull the character details in automatically. Lorebooks are first-class too, importable, exportable, and attachable to any character rather than locked to their creator's setup. The practical consequence: your investment in characters and world notes is not hostage to any one frontend. A card written for SillyTavern plays in Wyvern, and a Wyvern card exports back out. Combined with a gateway key, that makes the whole setup portable in both dimensions at once, characters move between frontends, and models swap underneath them, without redoing either half. One habit worth carrying over from the SillyTavern world: keep the card and lorebook stable during a long story and let the model's instruction-following do the work. Model swaps mid-story are safe because history travels with the chat; card rewrites mid-story cost more coherence than they add.

Choosing models for Wyvern sessions.

The catalog mix that fits roleplay maps cleanly onto Wyvern's per-chat model choice. deepseek-v4-flash is the community's default daily driver: natural dialogue, forward-moving scenes, and rates low enough that a heavy evening costs a small fraction of a cent per message. deepseek-v4-pro upgrades description density for campaigns worth pinning. kimi-k2.6 and glm-5.2 break same-voice fatigue in rotation, with GLM's long-horizon adherence earning it detail-heavy scenes despite its reasoning latency. claude-sonnet-4-6 remains the prose ceiling for safe-for-work storytelling, priced accordingly. Content policy, stated plainly: Wyvern's own terms govern the platform, and each model family's policy governs what that model will write; no custom endpoint changes either. The policies-compared page linked below covers the landscape factually, and the honest summary is that policy fit is a model-selection criterion like price or prose, not something a setup guide can configure away.

Pay-as-you-go · transparent per-model pricing

Selected models are priced below official list prices. Exact input, output, cache, and per-request prices are shown for each model.

Cost shape: what a BYOK platform actually bills.

Because Wyvern charges nothing itself, your entire cost is the metered key, and roleplay's cost shape is input-dominated: the frontend resends the visible chat history with every message, so a long session bills its accumulated context again and again. That makes two settings matter more than any subscription comparison. Keep the working context deliberate rather than maximal, summarizing older events into lorebook entries the way the context-length guide below describes. And match the model to the scene: rotation between a value daily driver and an expensive finale model cuts the bill by an order of magnitude against running the flagship always. The starting credit here covers the evaluation phase without a card: enough identical prompts across deepseek-v4-flash, kimi-k2.6, and claude-sonnet-4-6 to know which voice your story wants, with per-request token counts in the usage log so the steady-state cost is arithmetic before any top-up.

FAQ