Connect Hugging Face

≈ 3 min Extension Web app JWT

Note — 

On-device models from Hugging Face (the Local AI page) need no token: they download without signing in.

Before you start

• A Hugging Face account.

Used by

• Hugging Face Embeddings

Set it up

Use Hugging Face Embeddings in a workflow. The key is a saved credential.

1. Create a key. In your Hugging Face settings, open Access Tokens and click Create new token. Choose Fine-grained and tick Make calls to Inference Providers, or choose Read. Copy the token. It starts with hf_.

2. Save it in AWFlow. Open Credentials and click Add Credential (or + Add Credential in the node’s Authentication field), then fill the New Credential form. In the browser side panel the form comes in two steps: tap Simple Credential (or click Continue), then fill the rest.

   123456

   1. 1 Step 1:Keep it a simple credential Leave Integration App / Service on Select an integration..., or pick Simple Credential at the top of its list.
   2. 2 Step 2:Name it For example Hugging Face – personal. A Description is optional.
   3. 3 Step 3:Set Auth Type to JWT Open Auth Type and choose JWT.
   4. 4 Step 4:Paste the key Paste the key into JWT, exactly as copied.
   5. 5 Step 5:Check where it is saved Signed in, choose Local (this browser only) or Cloud (synced to your account) under Storage location. Without an account it is saved locally in this browser, as here.
   6. 6 Step 6:Click Create Credential Check: the credential is listed on the Credentials page, ready to pick in nodes.

   Numbered areas in the screenshot: 1. Integration App / Service: none (Simple Credential); 2. Name; 3. Auth Type: JWT; 4. JWT; 5. Where the secret is stored (here, locally in this browser); 6. Create Credential.

3. Pick it in the node. In Hugging Face Embeddings, choose the credential in Authentication and type the Model id.

Check it's connected — 

Connect Hugging Face Embeddings to a Local Knowledge node and index one short text with the Indexer: it finishes without an error.

Where your token lives

Where your token lives — 

The credential is stored where you choose when you create it:

• Local (this browser only): encrypted in this browser. The encryption key is kept on the same device, so turn on the local secret vault to lock it with a passphrase only you know.
• Cloud (synced to your account): sent over an encrypted connection and stored encrypted on AWFlow’s server, so it works on your other devices.

When a step runs, your browser sends it only to Hugging Face Inference (huggingface.co). AWFlow’s servers never relay the call.

Privacy & your data →

Common issues

• A 401, “invalid API key” or “Incorrect API key” error. The key was mistyped, revoked, or belongs to another account. Create a new key and update the credential (or the provider in Settings › Providers).
• A 429 or “quota” error. The account is out of credit or over its rate limit. Add credit or wait, then run again.
• An embeddings node fails with 401 while a chat node with the same key works. The credential is a Bearer Token. Embeddings nodes pass the value to the provider as it is, so use a JWT credential that holds only the key.

Related

Hugging Face Embeddings node

All connections

Create and manage credentials

Privacy & your data