Settings · LLM Switchboard

🔒 You choose where your keys live. By default they stay in this browser only (local storage) and travel with your requests through the proxy to make the model call — held in memory for that call, never persisted or logged on our server. Opt in to “Save to vault” below and they're stored server-side in your Authlee Secrets Vault — encrypted and tied to your account — so they follow you across browsers. Either way they're per-user and never shared; clear them anytime.

Provider keys

These three are wired — the router picks among them. Want more? Onboard any OpenAI-compatible provider on Connections → (Together, Fireworks, Cerebras, OpenRouter…): paste your key there and address it as model: "<provider>:<model>".

Groq API key

not set

Free at console.groq.com/keys ↗.

NVIDIA API key

not set

Free credits at build.nvidia.com ↗.

Cloudflare Workers AI key

not set

Workers AI is account-scoped — paste ACCOUNT_ID:TOKEN (e.g. 3b9…ef:cfut_…). Find your Account ID on the Cloudflare dashboard (Workers & Pages → right-side Account ID); create a token with the Workers AI permission at dash.cloudflare.com ↗. A bare token only works if it can list your account.

Decision model (optional)

Not a chat provider — Jev answers typed questions instead of generating text, so it sits outside the router's model pool and is never part of a fallback chain.

TypeSafe API key

not set

Powers paper & article triage and prompt classification. Early access — join the waitlist at typesafe.ai ↗, then take a key from console.typesafe.ai ↗.

Sandbox (optional)

Railway token cloud sandbox

not set

Delegates a temporary model service to Railway — the Run locally page uses it to spin up an Ollama sandbox in the cloud (a billable Railway service you tear down when done). Create one at railway.app/account/tokens ↗. Test connection verifies it against the Railway API — nothing is deployed.