Hugging Face

Summary

Hugging Face is a commercial ai-infrastructure platform that hosts machine learning models, datasets, and demo applications (Spaces) on a Git-based hub, alongside libraries for building, training, and deploying them. It runs primarily as a hosted SaaS hub, backed by client libraries and a CLI for scripting and agent access, with paid tiers covering dedicated inference endpoints and org-level access controls. It targets machine learning engineers and applied researchers building or shipping models, rather than security teams directly, though its resource-group permissions and audit-log endpoints suggest platform and access-management concerns for organizations running it at scale. Its own documentation lists it alongside adjacent tools like Transformers, Diffusers, and Gradio rather than naming external competitors, positioning it as the shared hub those tools build on.

Hugging Face is a platform providing pre-trained machine learning models, datasets, and tools for natural language processing and computer vision tasks.

Release history

1. docs update Aug 29, 2026 · issue 010

   Hugging Face CLI v1.29.0 adds bucket visibility controls, hardware discovery, and managed engine deployment with parallelism flags

   └──▷ TRY IT
   Before deploying, find valid hardware combos in your region and see live quota and pricing to pick the right instance.
   copy

   $ hf endpoints hardware --vendor aws --region eu-west-1

   Deploy a multi-GPU model with vLLM, explicitly spreading across all 8 accelerators so the API accepts the configuration.
   copy

   $ hf endpoints deploy gpt-oss-120b-vllm \
     --repo openai/gpt-oss-120b \
     --engine vllm --custom-image vllm/vllm-openai:v0.23.0 \
     --tensor-parallel-size 8

   Lock a bucket down to private to prevent unintended public access.
   copy

   $ hf buckets settings username/my-bucket --private

   • ›Adds hf buckets settings <bucket> --private and --public flags to switch a bucket between private and public visibility.
   • ›Adds hf endpoints hardware subcommand to list deployable hardware combinations with vendor, region, accelerator, instance type, instance size, memory, GPU memory, price per hour, quota, and status.
   • ›Adds --vendor, --region, --accelerator, and --instance-type filter flags to hf endpoints hardware to narrow results to the exact values you intend to pass to hf endpoints deploy.
   • ›Adds --all flag to hf endpoints hardware to include hardware that is deprecated, temporarily unavailable, or out of quota (hidden by default).
   • ›Adds --engine flag to hf endpoints deploy to run a container as a managed engine (vllm, sglang, tgi, tei, llamacpp, hf-serve, …), unlocking engine-specific settings.
   +2 moreshow less

   • ›Adds --tensor-parallel-size and --data-parallel-size flags to hf endpoints deploy and hf endpoints update, writing parallelism config into the engine's model.image field that the API validates against the instance's accelerator count.
   • ›Adds --engine flag to hf endpoints update as the only CLI path to change an existing endpoint's image, replacing model.image entirely rather than patching it.
   └──▷ BREAKING ON UPGRADE

   • !The API now rejects vLLM and SGLang deployments where --tensor-parallel-size and --data-parallel-size are both unset, because those engines default to one accelerator while the endpoint receives every accelerator of its instance.
2. 0.0.1 Aug 28, 2026 · issue 009

   API surface changed: +6 endpoints, 5 modified

   API CHANGE

   API surface changed: +6 endpoints, 5 modified

   • + GET /api/users/{username}/followers
   • + GET /api/users/{username}/following
   • + GET /api/users/{username}/following/orgs
   • + GET /api/{repoType}/{namespace}/{repo}/duplicate/status
   • + PATCH /api/collections/{namespace}/{slug}-{id}/items/{itemId}
   • + PATCH /api/collections/{namespace}/{slug}/items/{slug}
   • ~ GET /api/datasets/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ~ GET /api/models/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ~ GET /api/spaces/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ~ POST /api/repos/create: response schema changed
   • ~ POST /api/{repoType}/{namespace}/{repo}/duplicate: response schema changed
   • ›New endpoint GET /api/users/{username}/followers
   • ›New endpoint GET /api/users/{username}/following
   • ›New endpoint GET /api/users/{username}/following/orgs
   • ›New endpoint GET /api/{repoType}/{namespace}/{repo}/duplicate/status
   • ›New endpoint PATCH /api/collections/{namespace}/{slug}-{id}/items/{itemId}
   +6 moreshow less

   • ›New endpoint PATCH /api/collections/{namespace}/{slug}/items/{slug}
   • ›GET /api/datasets/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ›GET /api/models/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ›GET /api/spaces/{namespace}/{repo}/lfs-files: param direction (query) added; param sort (query) added
   • ›POST /api/repos/create: response schema changed
   • ›POST /api/{repoType}/{namespace}/{repo}/duplicate: response schema changed
3. 0.0.1 Aug 22, 2026 · issue 004

   API surface changed: 3 modified

   API CHANGE

   API surface changed: 3 modified

   • ~ GET /api/datasets/{namespace}/{repo}/user-access-request/{status}: response schema changed
   • ~ GET /api/models/{namespace}/{repo}/user-access-request/{status}: response schema changed
   • ~ GET /api/users/{username}/overview: response schema changed
   • ›GET /api/datasets/{namespace}/{repo}/user-access-request/{status}: response schema changed
   • ›GET /api/models/{namespace}/{repo}/user-access-request/{status}: response schema changed
   • ›GET /api/users/{username}/overview: response schema changed
4. 0.0.1 Aug 19, 2026 · issue 001

   Hugging Face now publishes an API — 316 endpoints across 27 areas: Spaces, Datasets, Models, …

   • ›Spaces (41 endpoints) — Get information from all Spaces on the Hub.
   • ›Datasets (36 endpoints) — Get information from all datasets on the Hub.
   • ›Models (36 endpoints) — Get information from all models on the Hub.
   • ›SCIM (28 endpoints) — Use the SCIM API to control and manage your hub Enterprise organization manage members' access. - Must be organization owner - Use Access token with write permission on organization - Organization must be Enterprise…
   • ›Jobs (23 endpoints) — The following endpoints manage jobs.
   +4 moreshow less

   • ›Discussions (20 endpoints) — The following endpoints manage discussions.
   • ›Collections (18 endpoints) — Use Collections to group repositories from the Hub (Models, Datasets, Spaces and Papers) on a dedicated page.
   • ›Organizations (16 endpoints) — The following endpoints let you interact with Hub Organizations and their members.
   • ›19 more areas: Buckets, Users, Agentic Provisioning, Resource groups, Service Accounts, Paper pages, Webhooks, Notifications, Kernels, OAuth, Documentation, Repositories, SQL Console, Inference Endpoints, Repository Search, Agents, Auth, Container Registry, Tokens
5. Build Spaces with AI Agents seen Aug 19, 2026 · issue 001

   Hugging Face Spaces now supports AI agent-driven Space creation for models, papers, or local folders.

   • ›New option on the Space creation page at huggingface.co/new-space to build a Space using an AI agent, generating a command you paste into your agent to build and iterate automatically.
   • ›AI agent Space builder supports targeting a model, paper, or local folder as the source for the generated Space.
6. Egress metrics for users and organizations seen Aug 19, 2026 · issue 001

   Hugging Face adds egress usage metrics to user and organization dashboards, with per-member breakdowns.

   • ›Users can now view their CDN egress usage directly in the Hugging Face dashboard.
   • ›Organization dashboards now include a per-user egress breakdown showing how much data each member consumes.
7. Filter Jobs by Label seen Aug 19, 2026 · issue 001

   Hugging Face Jobs pages now support filtering by label via clickable chips and free-form key=value input.

   • ›Adds label-based filtering to user and organization Jobs pages, with most-used labels surfaced as clickable chips showing job counts.
   • ›Supports free-form key=value label input to filter by any label, including those not shown as chips.
8. Granular Feature Access seen Aug 19, 2026 · issue 001

   Hugging Face adds per-resource-group feature access controls, enabling fine-grained permissions beyond org-wide roles.

   • ›Enables feature access controls at the resource group level, so permissions like Jobs, Inference Endpoints, and blog publishing can be scoped independently per group rather than only by organization role.
9. MCP Server Enhancements seen Aug 19, 2026 · issue 001

   Hugging Face MCP Server gains unified hf_fs tool and Sandbox execution environments for AI assistants.

   • ›Adds the hf_fs tool — a single interface covering repositories, storage, documentation, and papers, with built-in search, operable in just over 1,000 tokens.
   • ›Adds Sandboxes, giving assistants secure execution environments attached to buckets and repositories for dataset analysis, model training, and Space creation.