Introduction

In this guide we'll walk through the process of linking Cerb to Hugging Face. You'll be able to use Hugging Face's full API in Cerb automations.

Get a Hugging Face API key

Choose Access Tokens in the menu and then click the "Create new token" button in the top right.

Name the key (eg. cerb), select the proper token type and permissions for your uses, and click Create token.

Copy the token for use later.

Create the Hugging Face service in Cerb

Navigate to Search » Connected Services.
Click the (+) icon in the top right of the list.
Select Hugging Face.
Paste the access token you copied earlier in the API Key field.
Click the Create button.

Create an agent model

An agent model record holds one model's configuration – its provider, endpoint and credentials – so automations reference it by name instead of repeating a provider block.

The fields below are in the order the form presents them. API endpoint URL comes before Model because it feeds both Refresh and Test.

Navigate to Search » Agent Models.
Click the (+) icon in the top right of the list.
Set Provider to Hugging Face.
Leave API endpoint URL blank. Its placeholder reads (auto), and blank uses https://router.huggingface.co. Set it only for a proxy or a self-hosted endpoint.
Set Authentication to the connected account you created above. The field isn't marked required, because a local provider needs none – but a hosted one will fail to authenticate without it.
Click the refresh button beside Model to load the provider's live model list, and pick one.
Give the record a short Name – this is what automations will use. Colons aren't allowed, since the name is referenced as cerb:agent_model:<name>.
Click Test to verify the connection, then Create.

Refresh is on demand, and it never happens quietly. Until a refresh succeeds, the suggestions under Model are a hardcoded list rather than your account's – there's deliberately no silent fallback, so a failed fetch looks like one instead of looking like nothing happened. Refreshing requires an administrator; a non-admin gets a permission error rather than an empty list. The field is free text throughout, so a model id works the day it ships even if the list hasn't caught up.

Give the refresh a moment to land before you open the Model menu. Opening it too early shows the shipped suggestions rather than your account's models, and the menu keeps showing them until you close and reopen it. Those are real model ids, so there's nothing on screen to tell you apart from a list that loaded.

The capability fields are almost always yours to fill in on this provider. Cerb has no table of its own for Hugging Face, so Context window, Vision, Thinking and the four Ratings normally stay exactly as you set them. The context window matters most – compaction ratios are fractions of it.

The exception to watch for is a model id beginning gpt-. Cerb falls back to its built-in OpenAI table for those, which sets Vision to Yes and a context window sized for OpenAI's own hosting rather than for this one. If either appears by itself, check it against the model's documentation rather than assuming it was measured here – a wrong Vision setting offers workers an image attachment the model then rejects, which is the more visible of the two failures. This only happens when you pick from a refreshed list; the shipped suggestions carry no metadata at all, so before a refresh nothing is filled in on any provider.

Use the model in automations

Reference the model by the name you gave the record.

start:
  llm.chat/summarize:
    output: results
    inputs:
      model: hf-llama
      messages:
        0:
          role: user
          content: Summarize this conversation in one sentence.
  return:
    summary@key: results:content

commands:
  llm.chat:
    allow@bool: yes

The same record works with llm.agent: for tool-using conversations, and with llm.router: to pick between several models as data. An automation that doesn't name a model resolves a pool instead – a search across agent models rather than a named record. Omit a search entirely and you get every available model, in the priority order an admin set.

Embeddings

Hugging Face also generates vector embeddings. llm.embed: names its provider inline rather than through an agent model record, so this one doesn't use the record you just created:

start:
  llm.embed/vectorize:
    output: embedding
    inputs:
      llm:
        huggingface:
          model: sentence-transformers/all-MiniLM-L6-v2
      input: The quick brown fox

Resources

Reference: Agent models