Table of Contents

Use AI inside PDMShell

Select AI in the main toolbar to write, explain, or troubleshoot scripts. The ? button beside the AI settings gear opens this guide.

Connect your provider

  1. Open AI, then select the settings gear.
  2. Choose ChatGPT, Claude, or Ollama.
  3. Complete the connection steps below and choose a model.
  4. Select Done and enter your message.
Provider How to connect
ChatGPT Locate the installed Codex CLI if prompted, complete the browser sign-in, and choose a model. See ChatGPT setup.
Claude Enter your Claude API key, connect, and choose a model. You can save the key encrypted for your Windows account. See Claude API setup.
Ollama Start Ollama with an available chat model, connect to its local address, and choose a model. The default address is http://127.0.0.1:11434/. Cloud models require sign-in through Ollama.

Use the provider menu beside the settings gear to switch saved connections. Each provider remembers its selected model. Switching providers or models starts a new conversation.

Your provider account determines model access and usage charges. PDMShell licensing and vault permissions also apply when running commands.

Choose your view

Control What it does
AI in the main toolbar Opens or hides the AI pane.
Divider beside the pane Drag to adjust the split between AI and the editor.
Expand in the AI header Fills the main window's content area with AI.
Return to split view Restores the editor and the previous pane width.
Pop out Moves the conversation into a separate, resizable window.
Dock or the separate window's close button Returns the conversation to the editor.

Expanding, resizing, and docking preserve your conversation, unsent message, and script draft. Use Stop to cancel a response. Closing the AI pane also stops a pending response.

Ask for help

Select the session tab you want to work with. Log into its vault if your task needs PDM access, then describe what you want to do. For example, ask the AI to find parts whose filenames contain membrane in the current folder and its subfolders.

Enter sends your message. Shift+Enter adds a new line. Select Include script to attach the active script to your next message. The checkbox clears after a successful response.

You can select and copy reply text or use Copy entire conversation. Clear text clears the displayed transcript. New chat resets the conversation context and approvals granted for that chat.

Review and use a script

When a response includes a script, open Script draft and review its commands and file targets.

Action What it does
Use draft… Asks before replacing the active editor's script text. Save your current work first.
Run below the draft Runs the draft in the selected session. Read the console for results.
Stop Requests cancellation of the running work. Completed changes remain applied.

Review generated commands before running them. Check file paths, searches, workflow transitions, and any command that changes vault data. Use the command reference when you need to confirm a parameter. Command reference

Allow AI to use commands

In AI settings, enable Use command catalog and run in the selected session if you want AI to inspect or act on the connected vault. For Ollama, choose a model that supports tools.

When an approval appears, review the session and exact command. Allow once approves that call. You can remember an exact command for the current chat and session, or allow read-only tools. Use Deny to reject a request.

Approved commands use the selected tab's vault login and current folder. Their results appear in the console and are returned to the AI provider. Use Revoke command approvals to require approval again. Remembered read-only access can be turned off in AI settings.

Share only scripts and command results your organization permits sending to the selected provider. Keep passwords, API keys, and license keys out of chat. The information icon provides connection and privacy details.

Work with vault scripts

Open a vault script through New/Open > Vault script in standalone PDMShell. Use AI beside the editor to ask for changes. Use draft… updates the editable code. Review it, then choose Save to publish the code and script settings. Vault script guide

Troubleshooting

Issue What to try
Connection required Open the settings gear, check the provider connection, and choose a model.
Codex executable missing Install the Codex CLI, then use Browse in AI settings to locate it.
Claude key or billing error Check your key and API account. Claude setup
Ollama has no models Start Ollama, make a model available, and reconnect.
Model rejects requests Check model access and account usage limits with your provider.
Vault command fails Check the selected session's login, permissions, license, and console error.
Response stops or fails Reconnect if requested and ask again. Review completed operations before retrying.

For using PDMShell from an external AI client, follow the separate setup guides. External Codex · MCP server