Install Edge Studio
Create and activate a Python 3.11 environment, then install Edge Studio from the Python package for normal local development:
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install edge-studio
edge doctor
If you use uv:
uv venv --python 3.11 .venv
source .venv/bin/activate
uv pip install edge-studio
edge doctor
Use the source path when contributing to Edge Studio or testing a local checkout.
Requirements
| Requirement | Version |
|---|---|
| macOS | 14 or later |
| Hardware | Apple Silicon |
| Python | 3.11 recommended |
| Node.js | Required only for building or developing the Web UI |
Install from source
git clone https://github.com/AtomGradient/edge-studio.git
cd edge-studio
python3.11 -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e .
edge doctor
Use edge for model readiness checks, model fetch receipts, and local learning demos:
edge models where qwen3.5-9b-4bit --json
edge models fetch qwen3.5-9b-4bit --source auto
edge demo chat --model qwen3.5-9b-4bit --interactive
edge models fetch --source auto can select the best available preview download path from ModelScope, Hugging Face, or an HF mirror. The download is explicit and writes a receipt.
After [chat:ready], ask a few normal questions and exit with /exit. The first 9B model load can take tens of seconds.
After the base chat works, continue to the CLI learning demo to inspect the synthetic correction sample and compare base vs Neural Imprint restored answer hashes.
Launch the Web UI
The installed package exposes a single edge command. Start the local Studio UI
and API server with:
edge studio
Open:
http://127.0.0.1:18842
The server runs on localhost by default. Stop it with Ctrl+C.
For frontend development from a source checkout, run Vite separately and keep the backend server running:
npm --prefix frontend ci
npm --prefix frontend run dev
edge studio
Build a wheel locally
Use the release packaging script when you need the same shape as the published package:
./scripts/build_wheel.sh
The script builds the frontend, packages the backend resources, and writes the wheel under dist/.