Add initial implementation of AI agent with mouse and keyboard control features

This commit is contained in:
Showdown76 2025-05-19 00:48:14 +02:00
parent ed34ebca6a
commit 7e612c1af7
8 changed files with 320 additions and 0 deletions

16
.vscode/launch.json vendored Normal file
View File

@ -0,0 +1,16 @@
{
// Use IntelliSense to learn about possible attributes.
// Hover to view descriptions of existing attributes.
// For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
"version": "0.2.0",
"configurations": [
{
"name": "Python Debugger: Current File",
"type": "debugpy",
"request": "launch",
"program": "main.py",
"console": "integratedTerminal"
}
]
}

29
ai/compute.py Normal file
View File

@ -0,0 +1,29 @@
import pyautogui
from objects.inputs import MouseInput, KeyboardInput, ButtonType
def press_mouse(mouse_input: MouseInput) -> None:
"""Presses mouse buttons at the given position."""
x, y = mouse_input.x, mouse_input.y
button = mouse_input.click_type
if button == ButtonType.LEFT:
pyautogui.click(x, y, button='left')
elif button == ButtonType.DOUBLE_LEFT:
pyautogui.doubleClick(x, y)
elif button == ButtonType.RIGHT:
pyautogui.click(x, y, button='right')
elif button == ButtonType.MIDDLE:
pyautogui.click(x, y, button='middle')
def press_keyboard(keyboard_input: KeyboardInput) -> None:
"""Types the given sequence of keys."""
text = keyboard_input.text
if text:
pyautogui.typewrite(text)
if keyboard_input.press_enter:
pyautogui.press('enter')
def _execute(name, args):
if name == "click_button":
press_mouse(MouseInput(**args))
elif name == "type_text":
press_keyboard(KeyboardInput(**args))

73
ai/processor.py Normal file
View File

@ -0,0 +1,73 @@
import traceback
import json # new
import openai
from objects import aic
import ai.compute
class AIProcessor:
def __init__(self, api_key: str, model: str = "gpt-4.1"):
self.oai = openai.Client(api_key=api_key)
self.model = model
self.session = aic.Session(messages=[aic.Message(role="system", content=aic.SYSTEM_PROMPT)], model=model) # type: ignore
self._tools_map = { # local binding of python callables
"click_button": self._click_button,
"type_text": self._type_text,
}
# --------------------- tool implementations --------------------- #
def _click_button(self, x: int, y: int, click_type: str) -> str:
# TODO: integrate real mouse automation.
return f"Performed {click_type} click at ({x}, {y})."
def _type_text(self, text: str) -> str:
# TODO: integrate real typing automation.
return f'Typed text: "{text}"'
def _execute_tool(self, name: str, arguments: dict) -> str:
func = self._tools_map.get(name)
if not func:
return f"Unknown tool: {name}"
try:
return func(**arguments)
except Exception as e:
traceback.print_exc()
return f"Error executing {name}: {e}"
# -------------------------- main entry -------------------------- #
def process(self, prompt: str, img_data: str | bytes | None = None) -> str | list[dict]:
try:
self.session.messages.append(
aic.Message(role="user", content=prompt, image=img_data)
)
response = self.oai.chat.completions.create(
model=self.model,
messages=self.session.messages_dict(),
tools=aic.FUNCTIONS, # type: ignore
)
# return tool call requests if any
tool_calls = getattr(response.choices[0].message, "tool_calls", None)
if tool_calls:
for tc in tool_calls:
ai.compute._execute(
name=tc.function.name,
args=json.loads(tc.function.arguments)
)
return [
{
"name": tc.function.name,
"arguments": json.loads(tc.function.arguments),
}
for tc in tool_calls
]
# otherwise return final assistant content
print(f"Response: {json.dumps(response.to_dict(), indent=4)}") # debug
output_text: str = response.choices[0].message.content # type: ignore
self.session.messages.append(
aic.Message(role="assistant", content=output_text)
)
return output_text
except Exception as e:
traceback.print_exc()
return f"Error processing request: {str(e)}"

18
main.py Normal file
View File

@ -0,0 +1,18 @@
from dotenv import load_dotenv
import os
import ai.processor
import webserver.web
load_dotenv()
def main():
aip = ai.processor.AIProcessor(
api_key=os.getenv("OPENAI_API_KEY", ""),
model=os.getenv("OPENAI_MODEL", "gpt-4.1")
)
server = webserver.web.WebServerApp(aip)
server.run()
if __name__ == "__main__":
main()

97
objects/aic.py Normal file
View File

@ -0,0 +1,97 @@
from dataclasses import dataclass
from typing import Optional, Union
from typing_extensions import Literal
compatible_models = ("gpt-4.1", "gpt-4.1-mini", 'o4-mini', 'gpt-4.1-nano')
SYSTEM_PROMPT = """
You are CopeAI Windows Agent. You are currently controlling a Windows 11 machine. \
You are capable to see the screen, click buttons, type text, and interact with the system. \
You will use the functions provided. The resolution of the machine is 1920x1080. \
Your text response must indicate what you are doing."""
FUNCTIONS = [
{
"type": "function",
"function": {
"name": "click_button",
"description": "Click a button at the specified (x, y) position with the given click type.",
"parameters": {
"type": "object",
"properties": {
"x": {
"type": "integer",
"description": "The X coordinate of the button."
},
"y": {
"type": "integer",
"description": "The Y coordinate of the button."
},
"click_type": {
"type": "string",
"enum": ["left", "double_left", "middle", "right"],
"description": "The type of mouse click to perform."
}
},
"required": ["click_type", "x", "y"],
}
}
},
{
"type": "function",
"function": {
"name": "type_text",
"description": "Type the given text at the current cursor location.",
"parameters": {
"type": "object",
"properties": {
"text": {
"type": "string",
"description": "The text to type."
},
"press_enter": {
"type": "boolean",
"default": False,
"description": "Whether to press Enter after typing the text."
}
},
"required": ["text", "press_enter"],
}
}
}
]
@dataclass
class Message:
role: Literal['user', 'assistant', 'system', 'tool'] # + tool
content: str
image: Optional[Union[str, bytes]] = None
disable_image: bool = False
name: Optional[str] = None # new only for tool messages
def to_dict(self) -> dict:
base = {
"role": self.role,
"content": self.content
if (not self.image and not self.disable_image)
else [{"type": "text", "text": self.content},
{"type": "image_url", "image_url": {"url":f"data:image/png;base64,{self.image}"}}],
}
if self.role == "tool" and self.name: # include tool name if present
base["name"] = self.name
return base
@dataclass
class Session:
messages: list[Message]
model: str = "gpt-4.1"
def to_dict(self) -> dict:
return {
"messages": [message.to_dict() for message in self.messages],
"model": self.model
}
def messages_dict(self) -> list:
return [message.to_dict() for message in self.messages]

20
objects/inputs.py Normal file
View File

@ -0,0 +1,20 @@
from dataclasses import dataclass
from enum import Enum
class ButtonType(Enum):
LEFT = "left"
DOUBLE_LEFT = "double_left"
RIGHT = "right"
MIDDLE = "middle"
@dataclass
class KeyboardInput:
text: str
press_enter: bool = False
@dataclass
class MouseInput:
x: int
y: int
click_type: list[ButtonType]

7
requirements.txt Normal file
View File

@ -0,0 +1,7 @@
openai
requests
flask
python-dotenv
# libraries to control mouse+keyboard+see screen
pyautogui
pynput

60
webserver/web.py Normal file
View File

@ -0,0 +1,60 @@
from flask import Flask, request, jsonify
import os, ai.processor
from dotenv import load_dotenv
load_dotenv()
class WebServerApp:
def __init__(self, aip):
self.app = Flask(__name__)
self._register_routes()
self.aip: ai.processor.AIProcessor = aip
def _register_routes(self):
@self.app.route('/api/test')
def test():
return jsonify({"message": "Hello, World!"})
@self.app.route('/api/request', methods=['POST'])
def handle_request():
# sent as form-data
data = request.form.to_dict()
if not data:
return jsonify({"error": "No data provided"}), 400
# Process the data as needed
prompt = data.get('prompt', '')
if not prompt:
return jsonify({"error": "No prompt provided"}), 400
img_data = None
if 'img' in request.files:
img_file = request.files['img']
if img_file:
img_data = img_file.read()
else:
img_data = None
import base64
# Convert image data to base64 if provided
if img_data and isinstance(img_data, bytes):
img_data = base64.b64encode(img_data).decode('utf-8')
elif img_data and isinstance(img_data, str):
img_data = img_data.encode('utf-8')
response = self.aip.process(prompt, img_data)
return jsonify({"response": response}), 200
@self.app.route('/api/health')
def health():
return jsonify({"status": "healthy"})
def run(self, *args, **kwargs):
self.app.run(*args, **kwargs)
# Example usage:
# if __name__ == "__main__":
# server = WebServerApp()
# server.run()