AISH (AI Shell) -- AI Agent with Memories

Screenshot showing a black screen with white text labels, and LLM chat controls

This is a Python GUI (SDL) application. You can chat with AISH, an AI agent with memory. You can also open multiple basic LLM chats, or create text notes. All of this is on a scrollable infinite canvas.

Aish (אֵשׁ): The Hebrew word "Aish" translates to "fire, light, or flame" in English.

In Korean, "Aish (아이씨)" or sometimes just "Ai (아이)" is a common informal exclamation often used to express frustration, annoyance, or mild surprise, somewhat equivalent to "Oh no!", "Darn!", or "Ugh!" in English.

So, Prometheus and Dr. Faustus. It captures the promise and the peril of AI.

Keyboard Commands

Verified on macOS and Windows

Mac	Windows	Description
Cmd+Q	Alt+F4	Quit
Cmd+S	Ctrl+S	Save workspace to file named by --workspace command-line param (default aish_workspace.json). Saving the workspace also saves agent memories to agent_memories.json, and saves the agent's perception of past events to agent_percepts.json
Cmd+L	Ctrl+L	Load workspace from file named by --workspace command-line param (default aish_workspace.json)
Cmd+B	Ctrl+B	Add a new Label at the current cursor position
Cmd+T	Ctrl+T	Create a new TextArea (use it like a post-it note)
Cmd+N	Ctrl+N	Create a new GPT-4 chat
Cmd+G	Ctrl+G	Send messages to GPT-4 for chat completion when you have an LLM Chat control focused
Cmd+U	Ctrl+U	Add another "user" message text field when an LLMChatContainer is focused
Cmd+Delete	Ctrl+Backspace	Delete the currently selected chat message (inside a chat), or the current chat (if whole chat is focused). Also works for other controls

Verified only on macOS

Mac	Windows	Description
Cmd+R		Use Voice out to say some sample text
Cmd+Enter		Toggle active listening mode (speech to text transcription)
Cmd+V		Paste text from clipboard into focused TextArea
TAB		Focus next control
Shift+TAB		Focus previous control
Enter		Move focus down into children of currently focused control
Esc		If the currently focused control has selected text, then clear the selection. Otherwise, move focus up to parent container
Ctrl+TAB		Insert TAB char into currently focused TextArea
Arrow L/R/U/D		Move within TextArea
Option/Alt+Arrow L/R		Move to start of previous / next word
Cmd+L/R Arrow		Move to beginning/end of current line.
Mouse Wheel Up/Down		Scroll focused TextArea
Cmd+Up		Move to beginning of focused TextArea
Cmd+Down		Move to end of focused TextArea
Shift+Movement Command		Extend selection within TextArea
Delete		Remove previous char in TextArea
Other keys		Insert text char into focused TextArea

Running it

python aish3.py

Command-line parameters/settings:

python aish3.py --help

yields:

usage: aish3.py [-h] [--fullscreen] [--width WIDTH] [--height HEIGHT] [--voice-in] [--workspace WORKSPACE]

AISH window application.

options:
  -h, --help            show this help message and exit
  --fullscreen          run in fullscreen mode
  --width WIDTH         window width (default: 1450)
  --height HEIGHT       window height (default: 800)
  --voice-in            Enable voice input (default: False)
  --workspace WORKSPACE workspace file (default: aish_workspace.json)

Setting up required environment variables

If you want to make GPT API requests, you'll need to provide a couple of environment variables in a .env file:

OPENAI_ORGANIZATION = "..."
OPENAI_API_KEY="..."

To use speech-to-text transcription, you'll need an AssemblyAI API KEY

ASSEMBLYAI_API_KEY="..."

In order to use voice output (text to speech), you'll need Google Cloud credentials:

GOOGLE_APPLICATION_CREDENTIALS="..."

And finally, to use the wakeup phrase detection, you'll need a PicoVoice access key:

PICOVOICE_ACCESS_KEY="..."

If you get tired of entering System prompt text into LLM chats, you can set a default system prompt using:

DEFAULT_SYSTEM_PROMPT="..."

NOTES:

Requirements / Prerequisites:

Python 3

Python SDL2 libraries for macOS: brew install sdl2

PySDL2 Python wrapper pip install pysdl2

Build a stand-alone macOS application

You can build a standalone application by doing: make macos

Run it as: ./dist/aish3

This requires the pyinstaller Python module pip3 install pyinstaller

jawa0 / aish3

readme