The ShareBuilder Assistant is expanding its AI workflow ecosystem with a Connector that brings Hugging Face models into the Assistant Skills platform. With the Hugging Face Connector, media teams can choose models for specific jobs, generate and summarize text, classify information, create embeddings, develop image concepts, and connect those results to sales, marketing, content, research, and operational workflows.
About Hugging Face
Hugging Face is an AI platform and community built around models, datasets, and applications. Its Model Hub includes models from many authors for language, image, audio, and other machine-learning tasks, while Inference Providers make supported models available through a common API.
For media organizations, that range can support account research, content analysis, internal knowledge discovery, customer-request routing, sales preparation, creative ideation, and other AI-assisted work. The value is choice: teams can evaluate models for the job instead of treating every workflow as a fit for one general-purpose model.
This is the Hugging Face Connector
The Hugging Face Connector brings selected Hugging Face Hub and Inference Providers capabilities into Assistant Skills.
The current connector is action-based. It can find and inspect models, generate or summarize text, classify content, create embeddings, generate images, and make an authorized API request for advanced needs. It does not include Hugging Face triggers, so workflows begin from another event such as a CRM change, form submission, approved process step, schedule, or manual request.
Choose a model for the job
Find Models searches the Hugging Face Hub by criteria such as keyword, author, task, or tag. Get Model retrieves details about a selected repository, including its task, library, tags, activity, access restrictions, and model status.
These actions can help a team build and maintain a reviewed model list for tasks such as summarization, classification, embeddings, chat, or image generation. A Skill could gather candidates for evaluation, inspect a model before it is approved, or flag models that are gated, private, or unavailable.
Popularity is not proof that a model is accurate, safe, licensed for the intended use, or compatible with the required provider. Teams should review the model card, license, intended uses, limitations, evaluation results, provider availability, and access requirements before using a model in production.
Turn approved business context into useful drafts
Send Prompt sends a prompt or conversation to a supported chat model. Summarize Text condenses source material with a selected summarization model.
A Skill could use these actions to:
- prepare an internal account or opportunity summary from approved CRM context
- draft sales-call questions, handoff notes, or a first-pass creative brief
- condense research, reports, transcripts, or customer requests
- create a short management update from a longer operational record
- reorganize approved information into a format another team needs
Generated text may be incomplete, inaccurate, biased, or inconsistent. Summaries can also lose qualifiers or omit information when source text is truncated. Workflows should preserve the original source and require human review before using AI output for customer communication, pricing, contracts, regulated information, or other high-impact decisions.
Route information and improve discovery
Classify Text assigns model-defined labels and confidence scores to text. Depending on the selected model, this can help route inbound requests, organize content by topic, identify sentiment in approved feedback, or send uncertain cases to a person for review.
Create Embeddings turns text into a numeric vector for semantic workflows. Embeddings can help teams find related account notes, articles, campaign briefs, or internal knowledge even when the records use different words.
A production classification workflow should be tested against representative business examples and should define what happens when confidence is low. Embedding workflows should use a consistent model and preprocessing approach, store the model identifier with each vector, and plan for re-indexing when the model changes. The connector creates embeddings; a separate governed vector store or workflow step must handle storage, search, access, retention, and deletion.
Explore visual concepts with human review
Generate Image sends a prompt to a compatible text-to-image model and provider. The image is saved to Assistant Skills file storage so later steps can route it for review, collaboration, or approved delivery.
Media teams could use this action for internal mood boards, campaign-concept options, draft visual directions, presentation assets, or early creative exploration before a designer develops final work.
Provider and model capabilities vary, and not every parameter works with every combination. Generated images require brand, quality, rights, disclosure, and appropriateness review. They should not be treated as final advertiser creative without the organization’s normal approval process.
Connect specialized AI to operating workflows
Hugging Face becomes more useful when a carefully selected model is connected to the systems where teams already work. Assistant Skills can combine it with ShareBuilder CRM, email, Microsoft Teams, forms, documents, approvals, file storage, publishing tools, databases, and other connectors.
A workflow could begin when a customer request arrives, classify it, route it to the correct team, summarize the relevant details, and create follow-up work for a person to review.
Another Skill could create embeddings from approved account notes, use a connected vector store to retrieve related context, ask a selected model to prepare an internal briefing, and require seller approval before saving or sharing the result.
Extend the connector for advanced needs
Make API Call can send an authorized GET, POST, PUT, PATCH, or DELETE request to supported Hugging Face Hub or Inference Providers endpoints. It gives technically managed Skills a path to operations that do not yet have a guided action.
Availability depends on the endpoint, token permissions, repository access, model and provider support, credits, billing, and rate limits. Advanced calls should be tested by qualified administrators and treated as supported only after their authorization, payloads, outputs, limits, and failure handling have been validated.
Govern model choice, data, and cost
Hugging Face models come from many authors and can differ in license, training data, access rules, capabilities, providers, and pricing. Before rollout, workflow owners should establish an approved model list and document:
- the model and provider used for each task
- permitted data and any access restrictions
- required human review and confidence thresholds
- expected inference and storage costs
- test cases, fallback behavior, and change-management rules
Prompts and source material may be processed by Hugging Face and the selected Inference Provider. Sensitive or regulated information should only be sent when that use and the provider have been approved. Production Skills should also handle unavailable models, timeouts, rate limits, exhausted credits, access changes, and unexpected output without silently taking a consequential action.
Assistant Skills
Skills are the workflows you train the ShareBuilder Assistant to run for your team. Create a Skill and the Assistant executes playbooks, coordinates processes, updates systems, and communicates across platforms automatically.
Availability
The current Hugging Face Connector package supports eight actions: Find Models, Get Model, Send Prompt, Summarize Text, Classify Text, Create Embeddings, Generate Image, and Make API Call. It does not include triggers.
A Hugging Face User Access Token is required. A fine-grained token needs permission to use Inference Providers, and private or gated models require appropriate repository access. Hugging Face credits, provider pricing, model access, and usage limits apply.
The guided connector does not include datasets, Spaces, repository publishing, model upload, fine-tuning, dedicated endpoint management, speech, audio, video, translation, or other specialized inference tasks. Selected supported operations may be reachable through Make API Call after separate technical and permission validation.
Connector availability within Assistant Skills depends on ShareBuilders installation, testing, rollout, and customer enablement. Presence in the upstream connector library does not by itself confirm production readiness for a specific customer.
For more information
To learn more about ShareBuilder Assistant Skills or discuss using the Hugging Face Connector in your AI workflows, visit ShareBuilder Assistant Skills or contact ShareBuilders.
About ShareBuilders
ShareBuilders delivers a media-focused sales platform that includes Pricing, CRM, Data Enrichment, and workflow automation. Built specifically for media organizations, ShareBuilders helps teams manage pipelines, activity, pricing, and performance with secure cloud access and mobile tools designed for sellers on the go.
