Quant picker that gets it right
Lists the real GGUF options from the repo with boundary-aware matching — Q4_K_M never matches IQ4_K_M.
From Hugging Face repo to live endpoint
Deploy an AI model to a cloud GPU without touching the CLI.
Paste a Hugging Face repo or collection URL, pick the quantized file, choose Google Cloud or DigitalOcean — Forge measures the weights, picks the right GPU, deploys llama.cpp, and smoke-tests the endpoint.
Working tool
Desktop tool
No charge today. We will share final pricing, platform support, and delivery terms before you decide to buy.
ForgeProduct screenshots coming soon
Real-size GPU picksGPU chosen from measured download bytes, not guesses
Safety gatesTyped CONFIRM before spending; DESTROY stops billing
115 self-testsInput parsing, GPU sizing, and log redaction verified
Focused workflow
Everything you need to move the work forward
A practical tool for a specific operational job, with less setup and less clutter.
Two deploy targets
Google Cloud Run (L4 or RTX PRO 6000) or DigitalOcean GPU droplets (RTX 4000 Ada up to H200).
Training tab
Fine-tune with Unsloth QLoRA on a GPU box, merge straight to GGUF, then serve or destroy the box.
Token hygiene
HF tokens and API keys travel in process env vars only, are never persisted, and are redacted from every log line.
Standalone scripts
Download the corrected deploy scripts for terminal use — Google Cloud and DigitalOcean versions.
Job history
Past jobs kept in a local SQLite database with secrets redacted.
How it works
From task to trusted record in three steps
Pick
Paste a Hugging Face repo, model URL, or collection URL and choose the quant file.
Configure
Pick Google Cloud or DigitalOcean and confirm settings.
Deploy
Type CONFIRM; Forge deploys, smoke-tests, and shows the connection card.
Release status
Clear about what is ready
We are preparing checkout, signing, licensing, and downloads. The page will switch to purchase actions only after the customer artifacts are verified.
Desktop app
Working toolForge is a working tool with 115 passing self-tests covering input parsing, GPU sizing, redaction, and the training pipeline. It shells out to the hf, gcloud, and doctl CLIs, which you install and authenticate yourself.
Android app
Desktop toolForge is a desktop developer tool. No mobile version is planned.
Launch note
Coming soonCloud GPU time costs real money — Forge requires a typed CONFIRM before deploying and a typed DESTROY before deleting billable boxes. Droplets bill hourly until destroyed, even when powered off.
Product questions
What teams ask before they reserve
What does Forge deploy?
Forge deploys quantized GGUF models from Hugging Face as llama.cpp servers — either on Google Cloud Run or on a DigitalOcean GPU droplet — then smoke-tests the endpoint and shows you the connection details.
How does Forge keep cloud spending safe?
Deploying requires typing CONFIRM, destroying a billable box requires typing DESTROY, and every deploy summary warns that droplets bill hourly until destroyed, even when powered off.
Need this workflow in your business?
This app is in development. We can still scope a workflow around your team, systems, and requirements.