Skip to content

Latest commit

 

History

History
345 lines (225 loc) · 11.3 KB

README.md

File metadata and controls

345 lines (225 loc) · 11.3 KB

What is GCA?

Hi, this is an open source framework to build vertical AI agent. We just support many llms and new technologies like mcp. You can build your own vertical ai agent army in few commands with the stucturized API.

GPT Computer Assistant - Create intelligence for your products | Product Hunt . Static Badge . Static Badge

Made_with_python . pypi_downloads



Playground of GCA | NEW

With playground.gca.dev you are ready to test and create your own strategies for creating an Vertical AI Agent.

  • Playground sessions limited to 10 minute.
Playground



GPT Computer Assistant(GCA)

GCA is an AI agent framework designed to make computer use across Windows, macOS, and Ubuntu. GCA enables you to replace repetitive, small-logic-based tasks worker to an AI. There is an really important potential that we believe. Whether you’re a developer, analyst, or IT professional, GCA can empower you to accomplish more in less time.

Imagine this:

  • Extract the tech stacks of xxx Company | Sales Development Representer
  • Identify Relevant tables for Analysis for xxx | Data Analytics
  • Check the logs to find core cause of this incident | Technical Support Engineer
  • Making CloudFlare Security Settings | Security Specialist

These examples shows how GCA is realize the concept of Vertical AI Agents solutions that not only replicate human tasks, GCA also in the beyond of human speed at same cases.



How GCA Works?

GCA is a Python-based project that runs on multiple operating systems, including Windows, macOS, and Ubuntu. It integrates external concepts, like the Model Context Protocol (MCP), along with its own modules, to interact with and control a computer efficiently. The system performs both routine and advanced tasks by mimicking human-like actions and applying computational precision.

1. Human-like Actions:

GCA can replicate common user actions, such as:

  • Clicking: Interact with buttons or other UI elements.
  • Reading: Recognize and interpret text on the screen.
  • Scrolling: Navigate through documents or web pages.
  • Typing: Enter text into forms or other input fields.

2. Advanced Capabilities:

Through MCP and GCA’s own modules, it achieves tasks that go beyond standard human interaction, such as:

  • Updating dependencies of a project in seconds.
  • Analyzing entire database tables to locate specific data almost instantly.
  • Automating cloud security configurations with minimal input.




Prequisites

  • Python 3.10



Using GCA.dev Cloud

Installation

pip install gpt-computer-assistant

Single Instance:

from gpt_computer_assistant import cloud, Task, TypeVerifier

# Starting instance
instance = cloud.instance()


# Add task
instance.add_task(
    Task(
        "Extract the github star number of https://github.com/Upsonic/gpt-computer-assistant", 
        TypeVerifier("integer")
    )
)

# Run tasks and get results
results = instance.run()
print(result)

instance.close()
Cloud




Self-Hosted GCA Server

Docker

Pulling Image

  • If you are using ARM computer like M Chipset macbooks you should use ARM64 at the end.
docker pull upsonic/gca_docker_ubuntu:dev0-AMD64

Starting container

docker run -d -p 5901:5901 -p 7541:7541 upsonic/gca_docker_ubuntu:dev0-AMD64

LLM Settings&Using

from gpt_computer_assistant import docker

# Starting instance
instance = docker.instance("http://localhost:7541/")

# Connecting to OpenAI and Anthropic
instance.client.save_model("gpt-4o")
instance.client.save_openai_api_key("sk-**")
instance.client.save_anthropic_api_key("sk-**")

# Asking and getting result
result = instance.request("Extract the tech stacks of gpt-computer-assitant Company", "i want a list")
print(result)

instance.close()



Local

Installation

pip install 'gpt-computer-assistant[base]'
pip install 'gpt-computer-assistant[api]'

LLM Settings&Using

from gpt_computer_assistant import local

# Starting instance
instance = local.instance()

# Connecting to OpenAI and Anthropic
instance.client.save_model("gpt-4o")
instance.client.save_openai_api_key("sk-**")
instance.client.save_anthropic_api_key("sk-**")

# Asking and getting result
result = instance.request("Extract the tech stacks of gpt-computer-assitant Company", "i want a list")
print(result)

instance.close()




Adding Custom MCP Server to GCA

instance.client.add_mcp_server("websearch", "npx", ["-y", "@mzxrai/mcp-webresearch"])

Roadmap

Feature Status Target Release
Clear Chat History Completed Q2 2024
Long Audios Support (Split 20mb) Completed Q2 2024
Text Inputs Completed Q2 2024
Just Text Mode (Mute Speech) Completed Q2 2024
Added profiles (Different Chats) Completed Q2 2024
More Feedback About Assistant Status Completed Q2 2024
Local Model Vision and Text (With Ollama, and vision models) Completed Q2 2024
Our Customizable Agent Infrastructure Completed Q2 2024
Supporting Groq Models Completed Q2 2024
Adding Custom Tools Completed Q2 2024
Click on something on the screen (text and icon) Completed Q2 2024
New UI Completed Q2 2024
Native Applications, exe, dmg Completed Q3 2024
Collaborated Speaking Different Voice Models on long responses. Completed Q2 2024
Auto Stop Recording, when you complate talking Completed Q2 2024
Wakeup Word Completed Q2 2024
Continuously Conversations Completed Q2 2024
Adding more capability on device Completed Q2 2024
Local TTS Completed Q3 2024
Local STT Completed Q3 2024
Tray Menu Completed Q3 2024
New Line (Shift + Enter) Completed Q4 2024
Copy Pasting Text Compatibility Completed Q4 2024
Global Hotkey On the way Q3 2024
DeepFace Integration (Facial Recognition) Planned Q3 2024

Capabilities

At this time we have many infrastructure elements. We just aim to provide whole things that already in ChatGPT app.

Capability Status
Local LLM with Vision (Ollama) OK
Local text-to-speech OK
Local speech-to-text OK
Screen Read OK
Click to and Text or Icon in the screen OK
Move to and Text or Icon in the screen OK
Typing Something OK
Pressing to Any Key OK
Scrolling OK
Microphone OK
System Audio OK
Memory OK
Open and Close App OK
Open a URL OK
Clipboard OK
Search Engines OK
Writing and running Python OK
Writing and running SH OK
Using your Telegram Account OK
Knowledge Management OK
Add more tool ?

Predefined Agents

If you enable it your assistant will work with these teams:

Team Name Status
search_on_internet_and_report_team OK
generate_code_with_aim_team_ OK
Add your own one ?
Logo

Contributors