Downloading a model from Hugging Face may look straightforward. Large repositories, interrupted transfers, gated models, multiple revisions, duplicate files and several ComfyUI environments quickly make the process harder to control.
BC ModelVault was developed as a graphical Hugging Face downloader and local AI model manager for Windows.
Stable release 1.0.1 downloads models, datasets and Spaces from a Hugging Face URL or repository ID while keeping progress, integrity, revisions and file organisation visible. It is available as a portable Windows application and does not require Python to be installed.
Download BC ModelVault for Windows ↗ View releases and features.
Why we developed BC ModelVault.
Hugging Face provides powerful official tools. Its command-line client is effective for developers, scripts, CI pipelines and terminal users, but everyday management of a local collection requires more than downloading.
You need to know which models are already present, where they are stored, which revision is installed, whether a transfer is intact and where checkpoints, LoRAs, VAEs, ControlNet models or LLMs belong.
BC ModelVault does not replace the Hugging Face CLI in automated workflows. It provides a visual workflow for organising and controlling a local library.
Download models, datasets and Spaces.
The application accepts a Hugging Face URL or an author/repository identifier. Before transfer, it inspects repository contents, available disk space and Windows filename compatibility.
Read tokens can be configured for private or gated repositories. They are managed through huggingface_hub and excluded from project files, logs and exported diagnostic reports.
A queue that survives restarts
The persistent SQLite queue restores tasks after the application or computer restarts. Downloads can be paused, resumed, cancelled and recovered, with speed, progress, estimated time and history kept visible.
Concurrency and bandwidth control
Users can select one, two, four or eight concurrent workers and set an aggregate speed limit. Temporary failures can be retried using a bounded progressive delay.
Precise revisions and integrity checks.
A repository changes over time. BC ModelVault associates each download with a verified revision, compares the installed and available commits and never silently overwrites an existing version.
When Hugging Face provides a reliable Git LFS digest, the software verifies the SHA-256 hash and assigns an explicit state:
- verified when the file matches the expected hash;
- unverified when no reliable hash is available;
- mismatch when the content differs from the published value.
Unverified does not automatically mean corrupted; it means that there is not enough information for reliable cryptographic verification.
A local library for AI models and assets.
Downloaded content enters a searchable catalogue covering language models, checkpoints, LoRAs, VAEs, ControlNet and other resources. Each entry can include author, repository, licence, revision, date, size, format, tags, type and integrity status.
Automatic classification with human control
The system suggests an asset type and displays its confidence. Users can correct the classification manually. Uncertainty remains visible rather than being hidden behind an automatic decision.
Duplicate detection
Asset hashes identify identical files even when their names, folders or source repositories differ, helping reduce storage use and reveal models already available.
Guided installation into ComfyUI profiles.
Multiple ComfyUI profiles can be configured for checkpoints, LoRAs, VAEs and ControlNet. Before making changes, ModelVault prepares a plan showing files, categories, directories, conflicts and installation method.
Copy, hard link and symbolic link are supported. If a selected method cannot be used safely, the application can fall back to copying. The plan remains visible and verifiable.
BC ModelVault and the Hugging Face CLI.
Diagnostics designed to protect data.
Rotating, redacted logs avoid recording Hugging Face tokens. Support reports exclude credentials, reducing the risk of sharing sensitive information accidentally.
How to download and install a model.
- Paste a Hugging Face URL or
author/repositoryID. - Select the destination folder.
- Add a read token when required.
- Use Check package to inspect files, space and compatibility.
- Select Download package.
- Pause and resume the transfer when necessary.
The portable executable requires no Python installation. Every release includes SHA256SUMS.txt. Because the executable is not digitally signed, Windows SmartScreen may display a warning; compare its checksum with the published value.
Download the latest release ↗ View source code and documentation.
Running from source requires Windows and Python 3.10 or later.
python -m pip install -r requirements.txt
python -m bc_modelvault.appLicence and commercial use.
BC ModelVault is distributed under the PolyForm Noncommercial License 1.0.0. Non-commercial use, modification and redistribution are permitted while retaining the licence and notices; commercial use requires separate authorisation. Review the repository's LICENSE and NOTICE files before use.
Software born from a concrete problem.
BC ModelVault reflects Balzano Consulting's software approach: avoid unnecessary complexity and design a focused tool around a real need. The challenge was not merely downloading a file, but recovering, verifying, organising, deduplicating and installing large assets correctly.
Does a technical or business process involve too many manual steps?