How to get more information why the Code Completion AI Agent in Theia IDE is not updating the editor file with the LLM response?

[six42]

Hello,
I am completly new to Theia so excuse me if I have some basic questions.
First of all: thanks for providing this great tool.

My Problem:
I use the CodeCompletion AI Agent in Theia, In the "AI Agent History" window I see that it calls the LLM model and the model returns a valid answer in the expected format. But the answer is not added into the file in the editor.
I had already working code completions. But often it is not adding the text.
In the LLM model output view I can see after the LLM respose the sentence: Cancel requested
But no further explication why. I also see this "Cancel request" message for a succesful and accepted Code Completion answer. :(

Where and how can I see more log information for this part of the Theia IDE application to understand why the answer is not added.
I asked the Theia Chat Bot already for help and tried all recommendations but it didn't helped.
I run Theia on Windows. I downloaded the installation package:
Version 1.59.1
VS Code API Version: 1.97.2

What the bot proposed and what I tried in italic:

To investigate why the AI-generated completion is not being inserted into the editor, you can check Theia's logs at different levels:

1. Check Theia Backend Logs

If you're running Theia from a terminal, the backend logs should provide insights. Start Theia with DEBUG logging enabled:

THEIA_LOG_LEVEL=debug yarn theia start

or if using Docker:

docker run -e THEIA_LOG_LEVEL=debug <your-theia-image>

This should show detailed logs, including possible errors when processing AI completions.

I checked the output on Standard Out, but nothing is added at the time of a Code Completetion AI Agent run.

2. Check Theia Frontend Logs (Browser Console)

Since Theia runs in a web environment, you can inspect the frontend logs:

• Open the Developer Tools in your browser (F12 or right-click → Inspect).
• Go to the Console tab and check for errors or warnings related to the AI completion process.
• Go to the Network tab to inspect API requests sent to the LLM and responses.

I checked the console output, but nothing is added at the time of a Code Completetion AI Agent run.

3. Enable Plugin/Extension Logging

If the AI Agent is implemented as a Theia extension, check if it provides additional logging options:

• Open the "Output" view (View > Output).
• Select the relevant output channel (e.g., Theia AI Agent or LLM Requests).
• Look for errors or missing events related to inserting completions.

There is an Output for the configured LLM model. It prints the request and the answer from the LLM and actually it adds a line "Cancel requested" after the LLM output

4. Manually Trigger AI Completion

Try manually invoking the completion using the appropriate command:

• Open the Command Palette (Ctrl+Shift+P).
• Type and select "Trigger AI Completion" (or a related command).

That's what I do all the time to control when code completion happens. I just us keys CTRL-ALT-SPACE

5. Check Theia's Log Files (If Persisted)

Some Theia applications store logs in files. Check if your setup logs errors to a file (~/.theia/logs or a similar directory).

There is a folder logs with a subfolder for each start of the IDE but there is no further log written.

6. Debug The Extension Code

If you are developing the AI agent or have access to the source code, you can debug it by:

• Attaching a debugger to Theia’s backend.
• Adding console.log statements in the extension’s code.

This I have not done yet, I hope sombody has a better proposal to understand the reason.

[sdirix]

Hi @six42!

We integrate the autocomplete suggestions via monaco.languages.InlineCompletionsProvider into Monaco.

We should investigate there whether this integration and completion provider can be improved.

[six42]

Hi Stefan @sdirix,
Many thanks for your quick answer.
This helps me to debug my problem further.