Getting Started
Foundry is an enterprise-grade unified hub platform for AI models, datasets, and applications.
It is compatible with the HuggingFace Hub while adding features required for enterprise environments, including on-premises deployment, enhanced security, GPU computing resource management, and synthetic data generation.
What is Foundry?
Foundry is a platform that enables organizations to systematically manage and collaborate on AI/ML assets in enterprise environments.
| Area | Value |
|---|---|
| Flexible Deployment | Choose on-premises or cloud |
| Ecosystem Compatibility | Compatible with the huggingface_hub library and tooling |
| Enterprise Security | 2FA, CSRF protection, token rotation, RBAC |
| Unified Workflow | Single platform from data preparation to model training and deployment |
| GPU Computing | Reactors — distributed GPU cluster management |
| Data Quality | Data Forge — synthetic data generation for LLM fine-tuning |
Key Components
Repository System
A version-controlled system for storing and collaborating on AI/ML assets.
- Models — Trained AI model files (PyTorch, TensorFlow, ONNX, etc.)
- Datasets — Training/evaluation datasets (Parquet, JSON, CSV, etc.)
- Spaces — Demo applications (Gradio, Docker)
Supports Git-style branches, commits, tags, and Pull Request-based collaboration.
Organization Management
Organization features for team-based collaboration.
| Role | Permissions |
|---|---|
| Owner | Delete organization, all administrative functions |
| Admin | Member management, settings changes |
| Member | View repositories and contribute |
Members are added via email invitation, and repository permissions are automatically granted based on their organization role.
Technology Stack
| Layer | Technology |
|---|---|
| Frontend | React, TailwindCSS, Monaco Editor |
| Backend | Next.js, Drizzle ORM, BullMQ |
| Database | PostgreSQL |
| Storage | LakeFS (version control) + S3/MinIO (file storage) |
| Auth | JWT + 2FA (TOTP) + CSRF (Double Submit) |
Quick Start
1. Create a Repository
- Click the + New button on the dashboard
- Select a repository type (Model / Dataset / Space)
- Set the name and visibility, then create
2. Upload Files
Upload files via drag-and-drop in the web interface, or use the Python SDK.
from foundry_hub import FoundryApi
api = FoundryApi(token="your_access_token")
api.upload_file(
repo_id="org/model-name",
path_or_fileobj="./model.safetensors",
path_in_repo="model.safetensors",
)
3. HuggingFace-Compatible Usage
Change just one environment variable to use your existing code with Foundry as-is.
import os
os.environ["HF_ENDPOINT"] = "https://your-foundry-instance.example.com"
from huggingface_hub import hf_hub_download
model_path = hf_hub_download(repo_id="org/model-name", filename="model.safetensors")
4. Import from HuggingFace
Import existing HuggingFace Hub repositories into Foundry with a single click.
- + New → Enable the From HuggingFace toggle
- Search for and select the repository to import
- Set the Foundry repository name and visibility, then create
Imports are processed in the background, and a notification is sent upon completion.
Deployment Options
| Option | Suitable Environment |
|---|---|
| On-Premises | Finance, public sector, defense, and other sectors requiring data sovereignty |
| Private Cloud | General enterprise VPC |
| Hybrid | Core data on-premises + computing in the cloud |
System Requirements (Minimum Specifications)
| Item | Specification |
|---|---|
| CPU | 4 Cores |
| Memory | 16 GB |
| Storage | 100 GB SSD |
| OS | Linux (Ubuntu 22.04+), macOS |
Next Steps
- Repositories — File management, version control, Pull Requests
- Organizations — Team collaboration and permission management
- HuggingFace Import — Migrating existing assets
- Python SDK / CLI — Programmatic access
- Security — Authentication, encryption, and compliance