PromptRocket is an experimental autopilot extension for VSCode leveraging the power of OpenAI's latest function call API. It's not just about having a chat panel in your editor, but also letting it drive VSCode autonomously to reach given targets. Discover more details and updates on the PromptRocket GitHub repository.
What distinguishes it from other AI tools?
PromptRocket operates based on some core philosophies.
Using OpenAI's recent function call update, PromptRocket can interact smoothly with VSCode. PromptRocket is now empowered to make recursive action decisions to accomplish the task at hand. Functions currently tackled include context retrieval from the current file, project-wide file list, symbols, and diagnostics retrieval, direct output to the editor, and self-command execution in VSCode. Autonomous behavior is recommended to use GPT-4, GPT-3.5 will not understand clear enough about the context and function usage to perform chained tasks.
🚨 Caution: It might mess up your code or the entire project. Use responsibly.
Prompts can be fully customized with PromptRocket. Modification extends from the system message to the use of prompt engineering techniques, like custom system messages, step-by-step thinking, and few-shot prompts. All are designed to extract the most effective response from GPT models.
PromptRocket remembers the chat history, your text selections, and their context. This extension's primary purpose is to enhance brainstorming by incorporating ChatGPT's conversation format into VSCode.
No flashy distractions on the panel, that's a promise.
Here are the current functionalities of PromptRocket that interface with VSCode:
How to Use
You can use PromptRocket in two ways: chat mode and command mode.
Customizable Chat Mode
Chat mode allows you to initiate a blank chat or use a custom template. Some templates might require extra context for optimised operations. The intelligent text recognition of PromptRocket ensures precise responses and complete context comprehension. It can efficiently track your focus across different files or code sections, offering deeper insights.
For repeating tasks like writing code comments, providing code snippets, etc., message shortcuts allow you to type less. Salient features like Insert into editor can be combined with commands to direct the result into your editor.
A couple of additional features are also included for enhanced user experience.
Prompt Template Examples
In PromptRocket, you can design various command templates and employ Few-Shot Prompts through example conversations. While the default prompts for coding are provided by PromptRocket, you can create your own for general writing tasks.
Here's a basic example:
This is a sample few-shot prompt, it involves giving the AI a few examples of the desired responses, and bolstering it with positive feedback. This tends to keep GPT-3.5's responses on track effectively.
Messages utilize handlebar templates allowing you to include specifics from the editor, such as the context surrounding the cursor, selected text, and the current language. Here is the full list of tokens: