Blueprint for Windows

The open-source desktop app for running, calibrating, fine-tuning, and operating open LLMs on your own hardware. Stock Q4 quants are calibrated to web text, not your workload — Blueprint lets you re-calibrate against your own prompts and ship a quant that's typically 8–15% more accurate at the same file size.

Apache 2.0. Runs entirely on your hardware. We capture your email so we can send the link — newsletter is a separate, optional opt-in.

Download

v0.3.6

Email me the download link

We'll send a one-click download link to your inbox. The link is good for 7 days.

Run

One-click llama.cpp install. GGUF model pull from HuggingFace. OpenAI-compatible API on a port you choose.

Calibrate

Run llama-imatrix against your prompts, quantize with that imatrix, ship a custom Q4 that beats the stock pre-quant by 8–15% on YOUR workload.

Tune

LoRA / QLoRA fine-tuning on your data via a bundled Python sidecar — runs on a single 24 GB consumer GPU. LLMLingua prompt compression.

Operate

Live GPU / VRAM / RAM metrics. SSH-managed remote fleets with push-install. Windows Service / systemd supervision for production.

After install

First-launch flow takes about 5 minutes.

The app opens straight to a welcome screen. Pick a model, size it for your GPU, Blueprint pulls the runtime + weights and starts serving. When you're ready, the Calibrate tab walks you through the imatrix → eval → calibrated-quant pipeline against your own prompts. Help articles are linked from every card via the ? icon.

Read the calibration guide Browse all how-to articles Try the planner first

Source on github.com/inspireailab-admin/blueprint-app. Kernel CLI at inspireailab-admin/blueprint-cli.

Power user · prefer the CLI?

The Blueprint kernel CLI does the install / pull / serve loop without a GUI — same code path as the desktop app.

# macOS / Linux
curl -sSL https://llmblueprint.ai/install.sh | sh

# Windows (PowerShell)
iwr -useb https://llmblueprint.ai/install.ps1 | iex

Apache 2.0 · source on GitHub