Skip to main content

Complete and accept tasks with AI Command Center

When to use this guide​

Use AI Command Center to understand the current app, design a Page or flow, change configuration, diagnose a failure, or produce a candidate implementation. AI is the default work entry point, but its result still requires acceptance in the editor and runtime environment.

Before you start​

  • Open the correct app and save current Page or Graph drafts.
  • Define the goal, allowed scope, protected areas, and success criteria.
  • Assign a human owner before work involving real equipment, Safety, permissions, or production deployment.
  • Never send passwords, API keys, access tokens, or sensitive customer data in chat, attachments, or screenshots.

Steps​

Configure a Provider and model​

  1. Open System Settings → AI Settings.
  2. Use an available Qixin AI model. Accounts with BYOK access can also add a Provider or sign in to the Codex Provider.
  3. For an API Provider, choose a preset or OpenAI-compatible, enter its display name, endpoint, and API key, then save.
  4. Select Refresh models on that Provider. Use Refresh all models when every Provider needs an update.
  5. Confirm that the model card lists the input capability needed by the task. A Vision task with images requires an image-capable model.

Open and manage sessions​

  1. Select the AI icon in the editor toolbar, or press Cmd/Ctrl+I.
  2. Use fullscreen for focused work, or choose Dock as tab to inspect the chat beside a Page or Graph.
  3. Select New chat for an independent task. Return to an existing session when continuing the same goal so its context remains intact.
  4. Choose a model and an appropriate thinking level, then send the first message.
  5. You can switch sessions while AI runs; the task continues. Stop ends the selected session's active run and clears its queued messages.

Give precise context with @​

  1. Type @ in the prompt.
  2. Choose a current-app file or a Skill that matches the task.
  3. Reference only the directly relevant Page, Graph, script, Query, or configuration.
  4. In the prompt, explain how AI should use each reference rather than sending file names alone.

For example:

Use @ to select the current equipment Page and the Page-design Skill. Keep the existing theme and add axis status, alarms, and a reset area. Do not change Safety or device addresses. When finished, list changed files and check Page diagnostics.

Write a task that can be accepted​

Use this order:

  1. Goal: what the user should receive.
  2. Scope: which resources AI may read and change.
  3. Constraints: behavior, interfaces, and safety configuration that must stay unchanged.
  4. Acceptance: checks AI must run and the expected result.

Template:

Goal: add a manual-operation Page for the packaging station. Scope: change only the referenced Page and required Page Methods. Constraints: reuse existing Objects; do not change Graphs, device addresses, or Safety. Acceptance: no Page diagnostics; Start and Stop call only existing Object methods; list changed files, human checks, and incomplete items.

Accept the result​

  1. Read the completion summary, validation results, and incomplete items. Do not accept a task only because AI says it is complete.
  2. Review changed files under Task details, then inspect every diff in the Git panel.
  3. Open each changed Page, Graph, Object, or Settings section and inspect its structure and properties.
  4. Review Problems and run a Build, Run, or Debug check appropriate to the risk.
  5. Test normal, boundary, failure, and recovery scenarios.
  6. If something is wrong, send the exact observed and expected behavior to the same session and request a narrow correction.
  7. Commit, Push, or deploy only after acceptance passes.

What success looks like​

  • The session uses the correct app, Provider, model, and references.
  • AI clearly reports its change scope, checks, remaining issues, and required human decisions.
  • The Git diff contains only approved changes.
  • Editor diagnostics and the agreed Build, Run, or Debug checks pass.
  • Real behavior matches the acceptance criteria in the prompt.

Common issues​

The model list is empty​

Return to AI Settings, confirm that the Provider is enabled and saved, check sign-in or API key status, and refresh models. A failed Provider does not block other available Providers.

@ cannot find a file I just created​

Confirm in Resources or Files that the file is saved and refresh that view, then type @ and search again.

AI keeps expanding the scope​

Stop the task. Restate the allowed files, protected areas, and smallest acceptance target. Start a new chat if old goals are interfering.

AI is running, but I need to add information​

You can add a queued message. If the current direction is wrong, stop first and continue with one corrected, complete request.

The session finished, but I am not sure it can go live​

That is expected. Ask AI to list unverified assumptions, then have the Page, flow, device, safety, and delivery owners accept their areas.

How AI can help​

Good AI tasks include:

  • Explaining how the current Page, Graph, Objects, Queries, and Variables work together.
  • Producing candidate changes to Pages, flows, configuration, or custom code.
  • Inspecting Problems, Build output, and broken references, then proposing a minimal repair.
  • Drafting a test list, change summary, or commit message.

AI must not decide real device parameters, safety policy, production downtime, or credential authorization on your behalf.

What a person must confirm​

  • AI is reading and changing the correct app, files, and task scope.
  • Page, Graph, and Object semantics match the real site process.
  • Device addresses, motion settings, PLC, Vision, permissions, and Safety are accepted by their responsible owners.
  • High-impact actions such as delete, Discard, Deploy, Stop, Commit, and Push are explicitly authorized.
  • API keys, passwords, and tokens were entered only in dedicated fields and did not enter chat or Git.