Skip to content

Assistant

Jump to bottom
pyscripter edited this page Apr 13, 2025 · 10 revisions

The Assistant

You can access the Assistant from the context menu of the editor.

graphic

Suggest
Request from the Assistant a suggestion for code completion. This is only available when there is no selection in the editor.

The following three commands are only available when a part of the code is selected.

Explain
Ask the Assistant to add comments explaining the selected piece of code. The comments should help users to understand the logic and functionality of the code.

Fix bugs
Ask the Assistant to fix the bugs in the selected piece of code. The Assistant is also expected to explain the nature of the original issues and how they were resolved, using comments in the code.

Optimize
Ask the Assistant to fix optimize the selected piece of code. The optimizations should maintain the original functionality of the code.

Cancel
Cancel a pending request. In all the above cases an activity indicator will appear in the PyScripter Status Bar. You can also cancel a request by clicking on the activity indicator.

Settings
Shows and allows you to modify the Chat settings (see below).

Assistant settings

Assistant settings are automatically saved when PyScripter exits and restored when PyScripter starts, in a Json file named "Assistant Settings.json" in the same directory as PyScripter.ini.

The settings are similar to the Chat Settings with the following differences:

Endpoint
The base URL for accessing the LLM API. You shouldn't need to change the provided defaults (which for OpenAI, DeepSeek and Olama are different from chat):

Model
The model you want to use. Currently the Assistant works with the following models:

  • OpenAI model:
    gpt-3.5-turbo-instruct
    The reason is that this is the only OpenAI model that has been trained for Fill-in-the-middle (FIM) tasks. And FIM is important for code completion. Alternatively, you can use the newer models, such as gpt-4o. In that case the endpoint needs to be changed to https://api.openai.com/v1/chat/completions
  • Gemini models:
    gemini-2.0-flash, gemini-2.0-flash-lite or any newer Gemini model
  • Deepseek model:
    deepseek-chat
  • Grok:
    grok-3-latest
  • Ollama models:
    Can use any model that supports FIM, such as:
    • codellama:code and its variants, e.g. codellama:7b-code-q6_K will work
    • qwen2.5-coder (recommended)
    • deepseek-coder-v2
    • codegemma:code and its variants, e.g codegemma:7b-code

Wider choice of Assistant models will be provided in the future.

The Suggestion Window

All Assistant responses will appear inside a pop-up Suggestion Window:

graphic

You may edit the response before issuing one of the following commands

Commands and shortcuts:

Accept (Tab)
Insert the contents of the suggestion at the cursor position.

Accept Word (Ctrl+Right)
Accept only the first word of the suggestion.

Accept Line (Ctrl+Enter)
Accept only the first line of the suggestion.

Cancel (Esc)
Ignore the suggestion and return to the editor.

You can see video demos of using the PyScripter Assistant at this blog post.

paypal

Clone this wiki locally