Set up global settings

Steps

1. 1

   Step 1: Prepare the workspace

   From the SIDERIUS checkout, run this command with your own absolute project path in place of /absolute/project. If already initialized this way, keep that workspace; running init again only fills missing content.

   .venv/bin/python -m tools.candidate_console init --workspace /absolute/project

   Copy

   Register your task in the new workspace's settings/console.json. Initialization saves and selects the standard model profile but does not create or populate data/<task_id>/; prepare the task data folder yourself.

   You should see: Your project has siderius_workspace, and the command prints configuration and token locations.

2. 2

   Step 2: Enable settings and connect

   Ensure console.json contains "global_settings": {"browser_writable": true}. Initialization adds it to new configurations; add it yourself to an existing one. Check the configuration, start the server and connect on Settings in the browser.

   .venv/bin/python -m tools.candidate_console check --config /absolute/project/siderius_workspace/settings/console.json
   .venv/bin/python -m tools.candidate_console serve --config /absolute/project/siderius_workspace/settings/console.json

   Copy

   You should see: Settings is writable and identifies the intended server.

3. 3

   Step 3: Choose models and data

   Choose a profile under Models and inspect Show routes for each step's model. Under Data folder, choose an allowed root for the task, open the desired subfolder and select Use <folder>. All paths are on the server. Symbolic links are visible but cannot be selected.

   If a choice is missing, have the console operator register it under llm_profiles or data_roots in console.json. You can also use Import from this computer… to import a JSON model profile; after validation it is saved in the server's library/llm/.

   You should see: Your intended profile and data folder are selected; active values have not changed.

4. 4

   Step 4: Save and restart

   Select Save settings. After Saved; active after server restart, stop the console server, run the same serve command again and reconnect. Inspect Values in effect. If a concurrent save is reported, select Reload and choose again.

   For example, global profile A does not replace profile B already set for the task in console.json. After restart, that task still uses B. Inspect its origin instead of repeatedly saving.

   You should see: Values in effect shows the values and origins used after restart.

5. 5

   Step 5: Keep advice and export when needed

   Import from this computer… under Advice accepts command-line-format advice, including earlier exports. Choose a file or None, but remember that the global choice is stored only; workflows keep their own advice.

   Under Export a workflow, select the project, task and export name, then Export and Download the files. The server saves llm.json and workflow.py, plus advice.json when the workflow has advice. Exports use active values and exclude pending settings and credentials.

   You should see: You know whether advice affects a run and can download the active configuration.