Get started
SoduWork has three parts. You need all three to run AI work.
| Part | What it does | Where it runs |
|---|---|---|
| Hub | Accounts, tasks, plugins, models and the admin console | A computer or server you control |
| Client | The desktop workspace people use | Each user's Windows, macOS or Linux computer |
| Runner | Starts the harness (runtime) and does the work | Inside the client, or on a shared server for cloud runs |
The client connects to one Hub. It does not start a Hub for you.
Choose a path
| Path | Use it when | Database | Who can reach it |
|---|---|---|---|
| Run from source | You want to try SoduWork on one computer | SQLite | This computer only |
| Deploy with Docker | You want a single-server install for a small team | PostgreSQL 16 | This computer, or your team behind your own HTTPS proxy |
| Deploy for a team | You need your own domain, backups and production settings | PostgreSQL | Your team |
The first two are separate installs with separate data. Moving from one to the other does not copy data.
Steps on this page group
- Requirements and limits: check your computer first.
- Run from source: start a Hub and open the admin console.
- Connect the desktop client: install the client and sign in.
- Configure a model and a harness: give SoduWork an AI model to use.
- Run your first task: send a chat, then run a plugin.
Then, as needed:
What SoduWork does not include
- Model keys. You configure model services you are allowed to use.
- Most harnesses. Pi is built in. Others are installed when you need them.
Without a model, you can use the admin console and collaboration features, but AI work will not run.
harness harnesses runtime runtimes 底座