Code you can return to
Build software, save conversations and pick up your projects again. Copy a code block with a single click.
Run compatible AI models on your own computer. Write code, work with files and create images without paying for local tokens. Once models are downloaded, local work can run offline.
v1.4.2 · Windows / macOS / Ubuntu · RU / ENModels run on your own CPU and GPU.
Local chats, files and project history.
A free application. Open source code.
Explore the source, follow releases and help shape what comes next.
Stay with the task. Project context, files and your conversation with the model are always within reach.
Let’s start with a clear structure. Here is the hero component:
export default function Hero() {
return (
<main className="your-next-idea">
<h1>Create something yours.</h1>
<p>Built with freedom.</p>
</main>
);
}
Build software, save conversations and pick up your projects again. Copy a code block with a single click.
Connect the Qwen Image model pack and create artwork on your own computer. Inference stays local.
Attach documents and images to your conversation. Analysis capabilities depend on the model you choose.

From a project cover to a new visual direction. Describe the idea and let a dedicated image model do the computation.
Explore the model setupThe first-run check inspects CPU, RAM, NVIDIA VRAM and disk space. Memory estimates help choose a size; they do not guarantee speed.
The app checks your processor, RAM, GPU memory and free storage. You choose a suitable model and download it from the library.
The check runs in the installed app. This website does not scan your computer.Huihui Qwen3.8 Coder 27B · IQ3_S3. Starting point: 32 GB RAM and NVIDIA 16 GB VRAM.
Qwen Image 2.1 Turbo · Q5_K_M. The generator, text encoder and VAE download separately.
Qwen3.5 4B and 0.8B. Lower memory requirements, with more limited coding and tool use.
Download the build for your system, choose EN/RU and an SSD folder. Python and local engines are included.
The app checks your PC, shows the library and verifies downloads with SHA256. Interrupted downloads can resume.
Code, documents and images in one window. Download the separate image model set to generate artwork.
Keep your app, models and history in one SSD folder. After downloading the models, local work is available offline.
Performance depends on the computer. Stop the app before disconnecting an external SSD. Moving to a different physical PC has not yet been tested.Choose your OS and download the app. Python and engines are included; models download separately.
An installer or a portable workspace. Choose a folder and make it yours.
First time? How to installTested on Windows 11. Python, Node.js and engines included; models separate.
A local workspace for Apple Silicon. Read the preview requirements before installing.
First time? How to installmacOS 26.2+, M1+. Unsigned candidate; native Mac execution not yet tested.
A standalone installer, a DEB package or an archive. Choose your preferred format.
First time? How to installTested offline in a clean container without system Python. GPU needs a driver.
Download directly from the project’s official GitHub releases, without an account. Python and engines are included. Choose your OS below.
EXE is the recommended installer. ZIP is portable: extract it completely into a new folder and run FreedomGPT.exe.
Click “Download for Windows” above. Open the downloaded EXE file.
Select your language and a new empty SSD folder, then click Install.
Open FreedomGPT.exe or the shortcut. The app checks your PC; choose and download a model.
Open PowerShell in a download folder. Paste the entire block. It downloads this release from GitHub, checks SHA256 and starts it. Enter a new empty SSD folder when asked. This creates a portable installation; no separate Python is needed.
& {
$ErrorActionPreference = 'Stop'
$freedomInstallPath = Read-Host 'New empty SSD folder (for example D:\FreedomGPT)'
if ([string]::IsNullOrWhiteSpace($freedomInstallPath)) { throw 'Choose an empty folder.' }
$freedomInstallPath = [IO.Path]::GetFullPath($freedomInstallPath)
$freedomSetupPath = Join-Path (Get-Location) 'FreedomGPT-Setup-1.4.2.exe'
$freedomHash = 'c7862b0c5e90ebcf454b405fa84047c76c966cd2346e1dae25392574dcd4317a'
if (!(Test-Path -LiteralPath $freedomSetupPath) -or (Get-FileHash -LiteralPath $freedomSetupPath -Algorithm SHA256).Hash -ne $freedomHash) {
curl.exe -fL -C - --retry 4 -o $freedomSetupPath 'https://github.com/Leongrupp/freedom-gpt/releases/download/v1.4.2/FreedomGPT-Setup-1.4.2.exe'
if ($LASTEXITCODE -ne 0) { throw 'Download failed. Run again to resume.' }
}
if ((Get-FileHash -LiteralPath $freedomSetupPath -Algorithm SHA256).Hash -ne $freedomHash) { throw 'SHA256 mismatch. Installation stopped.' }
$freedomProcess = Start-Process -FilePath $freedomSetupPath -ArgumentList ('/extract="' + $freedomInstallPath + '"') -Wait -PassThru
if ($freedomProcess.ExitCode -ne 0) { throw 'Installation failed. Choose a new empty folder outside your user profile.' }
Start-Process -FilePath (Join-Path $freedomInstallPath 'FreedomGPT.exe')
}If one-click copy is unavailable: Select → Ctrl+C (Cmd+C on Mac).
The ZIP contains a complete application. This preview requires M1+ and macOS 26.2+, and is unsigned/not notarized.
The preview requires Apple Silicon M1+ and macOS 26.2+. Intel Mac is not supported.
Click “Download for Mac”, unpack the ZIP and move Freedom GPT.app into Applications.
Open Freedom GPT.app. The unsigned preview may require manual OS approval; native Mac execution is not yet verified.
Open macOS Terminal in a download folder. Paste the entire block. It downloads this release from GitHub, checks SHA256 and starts it. These commands have not been executed on a physical Mac. OS approval may still be required.
(
set -eu
[ "$(uname -m)" = arm64 ] || { echo 'Apple Silicon is required.'; exit 1; }
sw_vers -productVersion | awk -F. '{ exit !($1 > 26 || ($1 == 26 && $2 >= 2)) }' || { echo 'macOS26.2+ is required.'; exit 1; }
freedom_file='FreedomGPT-1.4.2-macOS-AppleSilicon.zip'
freedom_sha='361c4703cb45f5c44c8b9bd2a4a4cb9a817b1cb9380a16ab9fb52cd75caf1d71'
freedom_folder='./FreedomGPT-1.4.2-macOS'
[ ! -e "$freedom_folder" ] || { echo 'The extraction folder already exists. Choose a new folder.'; exit 1; }
if ! printf '%s %s\n' "$freedom_sha" "$freedom_file" | shasum -a 256 -c - >/dev/null 2>&1; then
curl -fL -C - --retry 4 -o "$freedom_file" 'https://github.com/Leongrupp/freedom-gpt/releases/download/v1.4.2/FreedomGPT-1.4.2-macOS-AppleSilicon.zip'
fi
printf '%s %s\n' "$freedom_sha" "$freedom_file" | shasum -a 256 -c -
ditto -x -k "$freedom_file" "$freedom_folder"
open "$freedom_folder/Freedom GPT.app"
)If one-click copy is unavailable: Select → Ctrl+C (Cmd+C on Mac).
.run is the recommended installer. DEB is for your package manager; TAR.GZ is an archive for manual installation.
Click “Download for Ubuntu”. The recommended file ends in .run.
Open Terminal in the folder containing the file. Copy and run the short command below.
Wait for the app to open. It checks your hardware and lets you choose a suitable model.
sh ./FreedomGPT-Setup-1.4.2-Ubuntu-x64.runIf one-click copy is unavailable: Select → Ctrl+C (Cmd+C on Mac).
Open Ubuntu Terminal in a download folder. Paste the entire block. It downloads this release from GitHub, checks SHA256 and starts it. If curl is missing, use the download button and the short command above.
(
set -eu
freedom_file='FreedomGPT-Setup-1.4.2-Ubuntu-x64.run'
freedom_sha='c6e4100867c74ee963f32cddbeb2ad3191562f791ee22e2cd64bace82e684910'
if ! printf '%s %s\n' "$freedom_sha" "$freedom_file" | sha256sum -c - >/dev/null 2>&1; then
curl -fL -C - --retry 4 -o "$freedom_file" 'https://github.com/Leongrupp/freedom-gpt/releases/download/v1.4.2/FreedomGPT-Setup-1.4.2-Ubuntu-x64.run'
fi
printf '%s %s\n' "$freedom_sha" "$freedom_file" | sha256sum -c -
sh "./$freedom_file"
)If one-click copy is unavailable: Select → Ctrl+C (Cmd+C on Mac).
After launch: let the app check your PC, select a model and download it. Models need extra disk space. Local chats work offline after the models are downloaded.
Check each build’s requirements above. Model weights download separately. For the standard coder, plan around 32 GB RAM, 16 GB VRAM and 35 GB+ SSD space. GPU drivers are installed separately.
Freedom GPT is a free application with MIT-licensed source code. Local inference does not require a paid API or a subscription. You provide the computer, disk space and electricity; model licenses are separate. Not every model fits every PC.
Freedom GPT is free. If it helps your work, an optional donation supports development. Donations do not unlock paid features.
Thank you for every idea, bug report and contribution 💛bc1qkkzx7rzx9qace3xvyphcdfy4d5up5k9gzyhdc6Yes. The application is free, its source code is MIT-licensed, and local inference has no subscription or per-token charge. You provide hardware, electricity and storage. Models have separate licenses.
Yes, for local chat and files after the models are downloaded. Search, websites, downloads and update checks need internet.
Automatically after a new signed release is published. Installation waits for idle tasks, checks signatures and hashes, preserves user data and backs up replaced app files. You can disable automatic updates.
No. Speed depends on hardware, model and context length. One tested long code answer on an RTX 4080 measured 100.85 tokens/s. This is a case-specific result, not a guarantee. Review code and run tests.
Application source is MIT. Bundled components retain their licenses and notices; model licenses are separate. The clean installer includes no personal history or credentials from the author.
I am developing the Freedom GPT agent, new model options and video generation. Video Studio is experimental; testing and improvements continue.