The CLI version of OpenAI's GPTs

[sigoden]

The CLI version of OpenAI's GPTs (Bots)

Isn't combining roles and function calls the essence of GPTs?

Why GPTs/Bots

Related functions are grouped together, making them suitable for specific tasks and easy to share.

File Structure

<llm-functions-dir>
├── bots
│   └── bot1
│       ├── embeddings                  # RAG files
│       ├── functions.json              # Function declarations
│       └── index.yaml                  # Bot definition
└── bots.txt                            # List of bots

<aichat-bots-dir>
└── bots
    └── bot1
        ├── message.md                # Saved messages
        ├── rag.bin                   # Generated embeddings store
        └── sessions/                 # Saved sessions

Bot definition

name: ''
description:  ''
version: ''
instructions: ''
variables:
    - name: ''
      description: ''
conversation_starters:
    - ''

CLI

--bot <name>          Select a Bot
--list-bot            List all installed Bots

Chat-REPL

.bot <name>           Enter Bot Mode
.info bot             View the Bot Info
.exit bot             Exit Bot Mode

Bot config

bots:
  - name: todo-sh
    model: openai:gpt-4o
    temperature: null
    top_p: null

Everyone is welcome to discuss what's missing and its value.

[blob42]

I think it would be great to have some sort of memory implementation for RAG on top of the knowledge ingestion.

I would imlement memory as a simple key,value with tag system.

knowledge.yaml

documents: (RAG ingestion)
 - path (dir or file)
 - ...

memory:
  location:
      value: (description of some location) 
      tags: birthplace,origin
  character:
      value: Foo is  (detailed character description ...)
      tags: main_character
  backstory:
      file: (path to file for ingestion)
      tags: ...
  

[sigoden]

#579 implment this feature.