Hugging Face: get an API key and write fiction with it on your phone

One HF token routes across a dozen-plus hosting partners (Groq, Fireworks, Cerebras, …) and a huge open-model catalog, with provider list prices passed through at zero markup; small monthly free credits keep light experimentation permanently free. This page is the BYOK quick answer for Hugging Face: key steps, direct console and pricing links, and the app-side notes that matter. Verified 2026-07-18.

Two links up front: get your key at the official console and check rates on the official pricing page. Prices and free quotas move; the official pages are authoritative.

What does it power inside Foreverse?

The factory preset declares 1 modality flag for Hugging Face. The models actually available come from the live catalog fetched in-app — catalogs move fast, and the flag only means the protocol wiring is ready.

Modality	Used for
Text	Continuation · roleplay chat · agent

How do I get a Hugging Face API key?

Sign up at huggingface.co and open Settings → Access Tokens.
Create a fine-grained token and tick the "Make calls to Inference Providers" permission — missing that box is the most common source of 401/403 errors.
Copy the hf_ token into the app.
Use Hub repo names as model IDs (like openai/gpt-oss-120b), optionally pinning a host with a :provider suffix (:groq, :fastest, etc.); by default routing is automatic.

Key in hand: three steps to connect

Open the app's Settings → AI models & services and find Hugging Face in the list (one of 62 presets, with logo and search).
Paste the key and tap the connectivity test — it fires one real request to verify key and network, and pulls back the model list while it's at it.
Enable the models you want, then set per-modality defaults (text, image, speech, and video each get their own); continuation, chat, and listening route through them from then on.

The one universal host pitfall: the API host must include its full version prefix (this provider's exact form is in the notes below) — the app never appends /v1 for you.

Hugging Face-specific notes

The preset host is https://router.huggingface.co/v1, speaking the Responses endpoint protocol (supported by the router, already wired in the app). Model names are Hub repo names, optionally suffixed :provider to pin a host; the default is auto-routing with failover, and the same model can be priced differently per host — the model page quotes are authoritative. The token must carry the "Make calls to Inference Providers" permission.

FAQ

Does Hugging Face have a free tier?

Yes: monthly free credits — $0.10/month on free accounts (officially "subject to change") and $2/month on PRO ($9/month); after that it's pay-as-you-go, or plug in your own provider key in HF settings (monthly credits then no longer apply). $0.10 covers only a few dozen small-model calls, so heavy use means PRO or top-ups.

Where does the key live once I paste it into the app?

It's stored on your device only, under system-level encryption — never uploaded or synced to any Foreverse server. Requests go straight from your phone to the provider's official endpoint with zero markup. To revoke access, delete the key in the provider's console.