GPT-Assistant(GUI)

Your friendly neighborhood GTK GUI chatbot, now with bells and whistles (functions).

Table of Contents

• Description
  • Limitations
  • TODO
• Functions
• Usage
• Screenshots
  • Chat UI
  • File-Manager
  • Settings
  • Read CSV and Plot with Python
  • Dev Flow

Description

Chatbot built with TKinter GUI and a bunch of functions.

• History is in practice in memory. (save to file with /save)
• File manager - View saved files, load text files into the chat.
• Has tiktoken calculation for total tokens/cost for both the input field and the total conversation.
• Requires usage of /commands to save on tokens from system messages and function params. /commands send the message to different agents based on the command with their specific system message and function params.

Limitations

• Pretty basic GUI - Not threaded, so it freezes when the bot is thinking.
• Can not copy from the chat window. (Workaround, ask it to write to file)
• I don't know python.
• Probably many more.

TODO / Wants / Would if I could

• Integration with Vector DB (A tab for it in the UI with upload/search/embed/etc)
• Settings tab for configuring the parameters of the different agents (somewhat done)
• Streamed responses live updated in UI. Not sure how to get around threading (or how to do it generally), especially for running python code.
• Refactor the UI code to be more modular and less spaghetti (somewhat done, tabs are seperated to their own files)
• Token counter / Cost counter
• Probably more

Functions

• /help - lists all available commands
• /csv - csv handler
• /python - python interpreter
• /kb - knowledge base handler [read, write, list]
• /history - history handler [read, write, list] - Useful if you only want to summarize history
• /write - file writer chooses between data/, data/code/ and data/code/projects depending on what it's asked to write
• /read - file reader
• /edit - file editor (takes line range, text)
• /wikidata - wikidata sparql handler
• /image - image to text captioner
• /scrape - web scraper
• /save 'filename' - saves entire conversation history to file in history folder

Coding Assistant commands

These are not functions per se, they run a model with a specific system message for each that returns a response in a strict format. See system.py for the system messages. Also, screenshots.

• /review - performs a review of the code following a strict response format
• /brainstorm [n] - returns a list of n ideas for the topic following a strict response format
• /write_spec - Writes a spec for the brainstorm following a strict response format
• /ticket - creates a ticket for the brainstorm/spec following a strict response format
• /write_tests - writes tests for the ticket / file from ticket
• /write_code - writes code for a file from ticket
• /write_files - writes all the files/subfolders from the ticket/spec into the projects folder.

Example Usecase: /read_file code.ts/py/rs/etc -> /review -> /edit_file code.ts/py/rs/etc

Usage

• Create a .env file in the root of the project:
  • OPENAI_API_KEY=your-api-key
  • PINECONE_API_KEY=your-api-key
  • PINECONE_ENVIRONMENT=your-environment
• pip install -r requirements.txt to install dependencies
• python main.py to run the chatbot

GUI Screenshots

Chat UI (With token and cost counters)

File-Manager

Settings Tab

Read CSV and Use Python to plot (older GUI)

Development Flow demo:

Brainstorm

Write Spec

Create Ticket