Run your first task
Goal: have an agent do a small job in a folder, then use a plugin.
Before you start: a model is configured and Pi shows Ready to dispatch (Configure a model and a harness (runtime)).
1. Prepare a folder
Create an empty folder for this test, for example ~/soduwork-test, and put a small CSV file in it named data.csv. Use made-up data.
2. Set up the chat
On the client home page, check the controls under the message box:
| Control | Set it to |
|---|---|
| Local / Cloud (computer icon) | Local: the work runs on this computer |
| Work location | Choose an existing folder and pick the folder you created. The agent reads and writes files there. Default workspace creates a separate folder for this chat instead |
| Access | Default access: routine actions run automatically and sensitive actions ask first. Read only only reads; Full access can also write files, run commands and use the network |
| Model | The model you configured |
| Runtime | Pi |
You can change the model and harness later in the same chat. The history and folder are kept.
3. Send a request
Write a Python script in this folder that reads data.csv, counts empty values in each column, and prints the result as a Markdown table. Then run it.
The chat shows what the agent does as it works: the files it reads, the commands it runs and the files it changes. If it needs a sensitive action, it asks you first.
When it finishes, the new files appear in the reply. Choose a file name to open it. To change something, reply in the same chat.
4. Use a plugin
A plugin packages a method so that the agent follows the same steps every time.
-
In the message box, type
/. A list of plugins opens. Keep typing to filter it. -
Choose Plugin Developer.
-
Send:
Turn the CSV check we just did into a plugin that takes a CSV file and reports empty values per column.
The plugin guides the agent to produce a plugin draft that you can check, install on this computer and run. Plugins you make stay on your computer until you choose to upload them to your organization.
To browse every plugin, open Plugins in the left bar.
5. Optional: run in the cloud
If your organization has a cloud runner, choose the computer icon and switch to Cloud before you send. The work runs on that server, and you can close the client while it runs. See How a cloud run works.
Check that it worked
- The reply lists the new script, and the folder contains it.
- In the admin console, Usage and cost shows requests for your model.
Next
- Write your own plugin: Plugins overview
- Try other harnesses: Agent harnesses
- Set up for a team: Deploy with Docker