3. Setup
📍 Where you are: Basics 3/3 — ① Federated learning and FedOps → ② Data and model → ③ Setup → next: your role’s track
🎯 What you do in this step: Run Agent Studio on your PC, sign in to Studio and the FedOps Console, and get the practice template. Both the Owner and the Client must do this, because training happens not on the server but on each person’s PC (Studio).
- 1) Check your system
- 2) Install Docker Desktop
- 3) Run Agent Studio
- 4) Accounts and sign-in
- 5) Get the practice template
- ✅ Checklist
- Next
1) Check your system
python3 --version # must be 3.10 ~ 3.12
- 16 GB memory recommended (the Studio container and a CPU LLM run together; the LLM is used only in the Model Use Track)
- 20 GB+ free disk
- Internet connection required (Docker images, packages, LLM download)
2) Install Docker Desktop
Install and start Docker Desktop. On Linux, Docker Engine + the compose plugin is enough.
docker version --format '{{.Server.Version}}'
docker compose version
Use a recent Docker. With an old version such as Docker 20.10, the
fedopscommand cannot stop an already running Studio (unknown flag: --timeout).
3) Run Agent Studio
python3 -m venv ~/.venvs/fedops && source ~/.venvs/fedops/bin/activate
pip install -U fedops
fedops run agent-studio
Success when Studio (http://localhost:24368) opens in your browser. The first time it takes a few minutes to pull the Docker image.
On an Intel Mac, if
pip install -U fedopsfails with anllvmlitebuild error, install withpip install --no-deps fedops.
To stop, run fedops stop agent-studio. Studio keeps code and local data under ~/fedops-workspace/. Do not delete this folder.
Why does Studio run on my PC? In federated learning, data must not leave your PC. So training, checks, and joining all happen in Studio on your PC, and the web (Console) only does things that need no data (publishing, approval, server operation).
4) Accounts and sign-in
Create an account with Create account on the FedOps Console sign-in screen. One FedOps account signs in to both the Console and Studio.
| Goal | Accounts needed |
|---|---|
| One role only | 1 |
| Practice both roles alone (Owner joins its own Task) | 1 |
| Split Owner and Client (experience request and approval) — this tutorial’s setup | 2 — one for the Owner, one for the Client |
Sign in in two places separately.
- On the Agent Studio start screen, check that FedOps server is
Main FedOps, then sign in with your FedOps account - Also sign in to the FedOps Console in your browser — Open Console at the top right of FedOps Web (https://ccl.gachon.ac.kr/fedops), or https://ccl.gachon.ac.kr/fedops/console
⚙️ FedOps server — which FedOps server to connect to, that is, where federated learning aggregation runs. To use another server — on-premises, AWS, Kakao Cloud, and so on — add it with Add server and that server’s address. Accounts and Tasks are separate on each server, so this tutorial keeps
Main FedOps.
Studio and the Console do not share sign-in. With two accounts, always check that Studio and the Console use the same account. Otherwise you get “my Task is not visible”.
5) Get the practice template
git clone https://github.com/gachon-CCLab/FedOps-hands-on-guide.git
cd FedOps-hands-on-guide/template
All later commands run from this template/ folder. Without git, use Code → Download ZIP on the GitHub page and unzip it.
| Role | What you use from the template |
|---|---|
| Owner | Almost everything — building per-hospital data, regular training code, Federated Task code, Agent rules, app code |
| Client | Only data/ — used to build the Cleveland hospital data (cleveland/dataset.csv). Studio downloads the code the Owner published. In real life a Client already has its own institution’s data and does not need the template |
✅ Checklist
docker compose versionprints a version- Agent Studio opens at http://localhost:24368
- Studio’s FedOps server is
Main FedOps - You are signed in to Studio and the Console with the same account
- The
template/folder is on your PC
Next
- Task Owner → Task Owner Track
- Client → Client Track