Hugging Face
AIReach thousands of open models through Hugging Face Inference Providers and the Hub API: chat, image and audio generation, embeddings, text analysis, and model discovery.
What This Integration Enables
Hugging Face is breadth. Through Inference Providers it runs chat, image, audio, embedding, and text models, and through the Hub API it discovers models and datasets. For a task where a specialized open model outperforms a general one, this is how an agent reaches it without hosting anything. Choice is the value and the responsibility: with thousands of models available, the flow needs to route the output and know when a person should check it. An orchestration layer picks up the model's result, decides what continues automatically, and escalates the uncertain cases. FlowRunner is built for that layer, so an open model becomes a dependable step rather than a science project.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Specialized Classification
A classification task is a poor fit for a general model but well served by a fine-tuned one on the Hub. The agent runs each document through that task-specific model, reads the output into structured fields, and routes confident results automatically. Low-confidence cases go to a person to label. The specialist model does what a generic one could not, and a person handles only the edges.
Multimodal Content Pipeline
A content flow needs several model types: embeddings for retrieval, an image model for assets, and a text model for copy. The agent reaches each through Hugging Face without standing up separate integrations. Generated assets and copy go to a person for selection before publishing, so the pipeline is fast and still supervised.
Model Evaluation in a Flow
A team wants to compare open models on their own data before committing. The agent runs the same inputs through several models via the Hub and records the outputs side by side. A person reviews the comparison and picks the model, turning an ad hoc evaluation into a repeatable, documented step.
Human-in-Loop Highlight
With thousands of models in reach, the safeguard is not the model choice but the review point. When a Hugging Face model returns a low-confidence result, or produces content headed for publication, FlowRunner routes it through a [human-in-the-loop](/concepts/human-in-the-loop/) step: the agent pauses, presents the output and the context, and sends it to the right person via Slack. They label, select, or approve. The open models handle the volume; a person owns the uncertain and the customer-facing cases.
Agent Capabilities
15 actionsChat
2- Chat Completion Generates a text response for a single prompt through the Hugging Face Inference Providers router (OpenAI-compatible), which routes the request to partner providers such as Groq, Together, Cerebras, Fireworks and more.
- Chat Completion (Advanced) Sends a fully custom chat completion request through the Hugging Face Inference Providers router (OpenAI-compatible) with a complete messages array (multi-turn conversations, multimodal image_url content parts for vision models), tool/function calling passthrough, structured outputs via a response format object, penalties and sampling parameters.
Images
1- Generate Image Generates an image from a text prompt using a text-to-image diffusion model (FLUX, Stable Diffusion and more) served by the Hugging Face HF Inference provider.
Audio
1- Transcribe Audio Transcribes an audio file into text using an automatic-speech-recognition model (e.g. OpenAI Whisper) served by the Hugging Face HF Inference provider.
Embeddings
1- Create Embeddings Converts one or more texts into embedding vectors using a feature-extraction model served by the Hugging Face HF Inference provider, useful for semantic search, RAG, clustering and similarity.
Text Transformation
2- Summarize Text Produces a shorter version of a document while preserving its important information, using a summarization model (e.g. facebook/bart-large-cnn) served by the Hugging Face HF Inference provider.
- Translate Text Translates text from one language to another using a translation model served by the Hugging Face HF Inference provider.
Text Analysis
4- Classify Text Assigns labels with confidence scores to a text using a text-classification model served by the Hugging Face HF Inference provider, for sentiment analysis, language detection, toxicity screening and similar tasks with model-defined label sets.
- Classify Text (Zero-Shot) Classifies a text against an arbitrary set of candidate labels you provide, no training required, using a zero-shot-classification model (e.g. facebook/bart-large-mnli) served by the Hugging Face HF Inference provider.
- Fill Mask Predicts the most likely words for a masked token in a sentence using a fill-mask model served by the Hugging Face HF Inference provider.
- Answer Question Extracts the answer to a question from a provided context text using an extractive question-answering model (e.g. deepset/roberta-base-squad2) served by the Hugging Face HF Inference provider.
Hub
3- Search Models Searches the Hugging Face Hub for models by name, author, task (pipeline tag) and serving provider.
- Get Model Info Retrieves the details of a Hugging Face Hub model by its ID, including pipeline tag, tags, downloads, likes, library, gated status and available siblings (files).
- Search Datasets Searches the Hugging Face Hub for datasets by name and author, with sorting by trending score, downloads, likes or recency and a result limit.
Account
1- Get Account Info Retrieves the profile associated with the configured Hugging Face access token via the whoami-v2 endpoint, user or organization name, full name, email, plan and token permission details.
Start building with Hugging Face
$100 in credits. No card required. Connect in minutes.