Explore the demo
Browse fictional case data and the relationship explorer. Live model processing and saved evidence require the local application.
Explore demo cases →Get started / Windows x64
Your first case starts here.
Choose the browser demo or install the live workspace. This guide tells you exactly what to download to run Qwen, graph memory and audio tools on your computer.
Run on your computer
Install the four tools below, download the project, then run one guided setup command. The current launcher supports Windows x64.
Show me the downloads ↓Four tools. Official download links.
Download Python ↗
Install a 64-bit Python 3.11+ interpreter. Enable “Add Python to PATH” if the installer offers it. This runs the application and setup scripts.
Download Git for Windows ↗
Install with command-line access. Git downloads the project source and lets you receive updates.
Download Docker Desktop ↗
Choose Windows AMD64. Complete its WSL/virtualization setup, reboot if requested, then open Docker Desktop and wait for the Linux engine to run. Windows requirements.
Download FFmpeg ↗
Under Windows builds, follow gyan.dev and choose a release essentials ZIP. Extract it and add the bin folder containing ffmpeg.exe and ffprobe.exe to your user PATH.
Reopen PowerShell after installation. Keep internet access available for setup. Qwen weights are about 1.28 GB; Python packages, PyTorch, the runtime and Docker images need additional storage. The supplied launcher uses CPU inference and does not require a dedicated GPU.
How do I add FFmpeg to PATH?
Search Windows for “Edit environment variables for your account.” Choose Path → Edit → New, add the extracted FFmpeg bin folder, and save. Open a new PowerShell window. Device restrictions may require help from your administrator.
Confirm the tools are ready.
Open PowerShell and run each command. Each should print version or service information. Docker must show a running server.
python --version git --version docker info ffmpeg -version ffprobe -version
Let setup fetch the models for you.
In PowerShell, open the folder where you want to keep the project, then paste these commands. If you already cloned it, run only the last command from that project folder.
git clone https://github.com/Aaditya252/Forensic-Atlas.git cd Forensic-Atlas powershell -ExecutionPolicy Bypass -File scripts/setup_local.ps1 -Start
Setup checks prerequisites, creates a project Python environment, installs dependencies, downloads Qwen and llama.cpp, downloads AASIST-L, starts the app and runs readiness checks. It stops if a step fails.
No manual model-file search, Qwen training, OpenAI API key or Node.js installation is needed. The launcher fetches the FalkorDB image through Docker. This is a guided developer setup, not a standalone desktop installer.
Ready means every check says PASS.
Open the local app, select Open Workspace → Hybrid agent hub, and look for LOCAL MODEL READY. Choose Generate new draft to invoke Qwen. A fresh installation includes fictional Northstar data; the author's saved cases and sample downloads are not included.
Open app on this computer ↗For audio: upload a non-sensitive test recording, save its metadata inspection in Media analysis, then run the experimental audio screen in Deepfake lab. Qwen summarizes verified tool context; AASIST-L provides research audio cues. Neither establishes an authenticity verdict. Visual deepfake detection and automatic transcription are not installed.
Start without reinstalling.
Open Docker Desktop, then run this from the project folder. Keep the folder in place to preserve stored paths and graph mounts.
powershell -ExecutionPolicy Bypass -File scripts/start_forensic_atlas.ps1
Find the next step.
A command is not recognized
Check that the tool is installed and available on PATH, then open a new PowerShell window. If Python opens the Store, install/configure the real interpreter first.
Docker cannot connect
Open Docker Desktop and complete any setup or reboot it requests. Wait for its Linux engine, then retry docker info.
A package or model download failed
Read the error, check internet access, disk capacity and network restrictions, then rerun setup. Do not use a different model file to bypass a checksum mismatch.
Qwen or the application is unavailable
Run the startup command. Inspect output/llm.stderr.log for the model and output/api.stderr.log for the app. Run .\.venv\Scripts\python.exe scripts/deployment_preflight.py for health checks.
I still see demo data online
The public Vercel site is the demo. Use http://127.0.0.1:8100/ on the same computer where you installed the full application. Local links do not connect visitors to the author's laptop.