QuintX OS — Neural Automation & Vision System

Core Capabilities

Everything in one panel.

🌐

Web Search & Browse

Search Google, open URLs, and scrape full webpage content using headless Playwright — all from plain English commands.

v1 · Active

📄

File Management

Read, write, create and list files and folders on your PC. Summarise documents, save outputs, explore directories.

v1 · Active

🎙

Voice Commands

Click the mic button, speak your command, click again to stop. QuintX transcribes and executes immediately. No wake word.

v1 · Active

🖼

Image Analysis

Upload any image — screenshot, photo, diagram — and ask QuintX to analyse it, extract text, or answer questions.

v1 · Active

🤖

WhatsApp & Telegram

Reply to WhatsApp and Telegram messages automatically while QuintX is running. Custom agent with topic filtering.

v1 · Active

🔧

Tools Manager

Toggle any tool on or off. Grant explicit permissions for each capability. Add API keys per tool. Full control.

v1 · Active

💬

Built-in Guide LLM

No API key? QuintX ships with its own instant guide model — explains every feature, answers setup questions, fixes errors.

v1 · Active

🔒

Delete Guard

Any delete, erase, format, or wipe command triggers a confirmation screen. You must type YES DELETE exactly. No accidents.

v1 · Active

👁️

Screen Vision

AI captures your screen to RAM and analyses it. Never written to disk. Ask QuintX what's on screen anytime.

v2 · Coming

Architecture

Modules & Sub-modules

QuintX is organised into four clean layers. Each module has a single responsibility and can be updated independently.

🧠

Core Layer

core/

engine.py

Multi-provider LangChain agent. Initialises Ollama / OpenAI / Anthropic LLM, binds all tools, builds tool-calling agent with rolling 20-message history.

quintx_llm.py

Built-in guide LLM. 13 knowledge categories, regex keyword matching, instant responses — zero API, zero internet, zero latency.

installer.py

Auto pip installer with animated progress bar. Runs before any imports — uses stdlib only so it works before packages exist.

config.py

.env file manager. Reads and writes AI_PROVIDER, API_KEY, AI_MODEL. Handles first-run terminal setup wizard.

voice.py

Speech recognition (Google API) + pyttsx3 TTS. Non-blocking — runs in daemon thread. Gracefully fails if PyAudio missing.

🖥

UI Layer

ui/quintx_app.py

SideStrip

Always-visible icon bar. Independent PyQt5 window — can never be hidden by other panels. Q logo, nav icons (💬📋🔧👤🤖⇄✕).

ChatBar

Glass pill input bar at bottom-centre. Draggable anywhere. Image picker, mic toggle (start/stop), text entry, send button. Slides up/down with animation.

ResultPanel

WhatsApp-style chat — your messages right (green), AI left (dark). Live tool step display. Animated thinking dots. Full scroll, selectable text.

ToolsPanel

Categorised tool cards with ON/OFF toggles. Permission dialog on enable. API key field per tool. FREE/PAID/v2 badges.

BotsPanel

3-tab panel: WhatsApp config, Telegram config, Agent settings. START/STOP toggle with live ● ONLINE status indicator.

ProfilePanel

Provider switcher (Ollama/OpenAI/Anthropic), API key field, model name field. Saves to .env on click.

🔧

Tools Layer

tools/

system_tool.py

read_file — reads any local file. write_file — creates/overwrites files. list_directory — lists folder contents. All @tool decorated for LangChain.

browser_tool.py

search_web — Google search in browser. open_url — opens any URL. scrape_page — headless Playwright extracts full webpage text.

tools_config.json

Central config for all 12 tools. Stores enabled state, permissions, API keys, cost, category, and version (v1/v2). Auto-created on first run.

🤖

Bots Layer

bots/

bot_engine.py

Shared AI brain. Priority: pre-set Q&A → topic filter → QuintX AI → escalation. Configurable business context, allowed topics, response length.

whatsapp_bot.py

Flask webhook server on port 5000. Receives Twilio webhook, processes via bot_engine, replies via Twilio API. Runs as daemon thread.

telegram_bot.py

python-telegram-bot polling. Handles /start /help /status commands + all text messages. Runs as daemon thread, stop/start via BotsPanel.

bot_config.json

Stores Twilio + Telegram credentials, agent settings, pre-set Q&A, allowed topics. In .gitignore — never pushed to GitHub.

core/engine.py — Multi-provider AI initialisation

# QuintX Engine — same interface regardless of provider
class QuintXEngine:
    def _build_llm(self):
        if self.provider == "OLLAMA":
            from langchain_ollama import ChatOllama
            return ChatOllama(model=self.model, temperature=0)

        if self.provider == "ANTHROPIC":
            from langchain_anthropic import ChatAnthropic
            return ChatAnthropic(anthropic_api_key=self.api_key)

        from langchain_openai import ChatOpenAI   # default + Groq
        return ChatOpenAI(api_key=self.api_key, model=self.model)
      

Data Flow

How QuintX processes
your requests

You type or speak

Text goes into ChatBar. Voice uses the mic toggle — click once to start, click again to stop. Image attaches with the 🖼 button.

ChatBar · send_signal

Destructive keyword check

If your message contains delete, erase, format, wipe, rmdir — a confirmation dialog appears. You must type YES DELETE exactly before anything executes.

DeleteConfirmDialog · mandatory

AIThread starts in background

A QThread runs the AI so the UI never freezes. StepCapture intercepts stdout to capture every tool call, thought, and observation as it happens.

AIThread · QThread · StepCapture

LangChain agent selects tools

The agent reasons about your request and selects which tools to call. It can chain multiple tools — search the web, then save results to a file, in one command.

create_tool_calling_agent · LangChain 0.2+

Steps appear live in ResultPanel

Every tool call emits a step signal. You see → Action, 🔍 search_web, 👁 Observation in real time — just like Claude's internal view.

step signal · live display

Final answer as chat bubble

Done signal emits the complete response. It appears as a left-aligned AI bubble in the ResultPanel. Text is selectable and the panel scrolls to the latest message.

done signal · ResultPanel · auto-scroll

Tools System

12 tools. Every one
permission-controlled.

Every tool declares what system access it needs. You must tick every permission before it activates. Nothing ever runs without your explicit consent.

PERMISSION TYPES

🌐 Open browser & websites

📂 Read files from PC

💾 Write & modify files

🖥️ Capture screen (RAM only) [v2]

📬 Read & send emails [v2]

💻 Execute code [v2]

🌐

Web Search

Search Google

ONFREE

🔗

Open URL

Open any site

ONFREE

🕷

Web Scraper

Extract page text

ONFREE

📄

Read File

Read local files

ONFREE

✍️

Write File

Create & save files

ONFREE

📁

List Folder

Explore directories

ONFREE

👁️

Screen Vision

AI reads screen

v2PAID

📧

Email Agent

Read & send email

v2FREE

🤖

App Control

Control desktop apps

v2FREE

⚡

Code Runner

Execute scripts

v2FREE

🎨

Image Gen

DALL-E generation

v2PAID

🔌

Custom Slot

Your own tool

v2FREE

tools/system_tool.py — Adding a tool

from langchain_core.tools import tool

@tool
def read_file(file_path: str) -> str:
    """
    Read and return the contents of a local file.
    Use this when the user asks to read, open, or summarise a file.
    """
    with open(file_path, "r", encoding="utf-8") as f:
        return f.read(5000)  # first 5000 chars
      

How the Agent Works

The AI brain that picks
the right tool every time

QuintX uses LangChain's tool-calling agent. It reads your message, reasons about what tools to use, calls them in sequence, and synthesises the final answer.

AGENT PROCESSING PIPELINE

📥

User Input Received

Text, voice transcript, or text + image path arrives at the engine. Full conversation history (last 20 messages) passed as context.

chat_history · rolling window

🤔

LLM Reasons & Plans

The system prompt tells the LLM its name, capabilities, available tools, and rules. It decides which tools to call and in what order.

system prompt · create_tool_calling_agent

⚡

Tool Called with Arguments

Agent calls the selected tool with the exact arguments. StepCapture intercepts stdout so you see every step live in the UI.

@tool · StepCapture · step signal

👁

Observation Fed Back

Tool result returned to agent as an Observation. Agent may call another tool (up to 6 iterations) or proceed to final answer.

max_iterations=6 · handle_parsing_errors

✅

Final Answer Displayed

Done signal emits the complete response. Shown as AI bubble in ResultPanel. History updated for next message.

done signal · AIMessage · history

core/engine.py — Agent construction

from langchain.agents import create_tool_calling_agent, AgentExecutor
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder

prompt = ChatPromptTemplate.from_messages([
    ("system", "You are QuintX, a desktop AI agent. Use tools to help the user."),
    MessagesPlaceholder("chat_history"),   # rolling 20-message memory
    ("human", "{input}"),
    MessagesPlaceholder("agent_scratchpad"), # tool call reasoning
])

agent   = create_tool_calling_agent(llm, tools, prompt)
executor = AgentExecutor(
    agent=agent, tools=tools,
    verbose=True,             # steps captured by StepCapture
    handle_parsing_errors=True,
    max_iterations=6
)
      

Messaging Bots

Reply on WhatsApp &
Telegram automatically

While QuintX is running on your PC, it can reply to your customers on WhatsApp and Telegram with a custom business agent.

💬 WhatsApp Bot

via Twilio

Uses Twilio's WhatsApp API. A Flask webhook runs on port 5000 on your PC. Use ngrok to expose it publicly.

In QuintX → click 🤖 → WhatsApp tab → paste credentials → Save

Run ngrok http 5000 → copy the https URL

In Twilio → set webhook to https://YOUR_NGROK/whatsapp

Click START → status shows ● ONLINE

✈ Telegram Bot

Completely Free

Uses python-telegram-bot polling. No webhook or ngrok needed. Goes live in seconds after pasting your token.

Open Telegram → search @BotFather

Send /newbot → enter a name and username

Copy the BOT TOKEN from BotFather's reply

In QuintX → click 🤖 → Telegram tab → paste token → Save

Click START → bot goes live instantly ● ONLINE

Agent Configuration — 🤖 → Agent Tab

What you configure

Business Name & Context
Tell the bot who you are and what you do

Allowed Topics
Bot only answers questions on these topics — everything else gets the off-topic reply

Pre-set Q&A
Instant answers for common questions — no AI needed, zero latency

Escalation Message
Shown when bot can't answer — redirects to email or support

Bot response priority

# 1. Pre-set Q&A (instant)
if keyword_match(msg, triggers):
    return preset_answer   # zero latency

# 2. Topic filter
if not on_topic(msg):
    return off_topic_response

# 3. AI engine
if self.engine:
    reply = self.engine.run(msg)
    return trim(reply, max_len)

# 4. Escalation
return escalation_message
          

AI Providers

Your AI. Your choice.

Switch providers anytime from 👤 Settings. No code changes — just update your .env.

🖥 Ollama

LOCAL · SLOW

mistral:latest · qwen3:8b · llava

Runs entirely on your PC. No internet. No API key. Complete privacy. QuintX scans installed models and shows a clickable list in the setup wizard.

Cost: Free forever · ollama.ai

⚡ Groq

CLOUD · FASTEST

llama-3.1-70b-versatile

The fastest option available — custom LPU hardware makes it faster than typing. Generous free tier. Recommended for most users wanting fast responses.

Cost: Free tier · console.groq.com

◆ OpenAI

CLOUD · VERY FAST

gpt-4o · gpt-4o-mini

Industry standard. GPT-4o includes vision built-in so image analysis works immediately. gpt-4o-mini is 10× cheaper for simpler tasks.

Cost: ~$0.01–0.05 per conversation

◇ Anthropic

CLOUD · VERY FAST

claude-3-5-sonnet-latest

Claude excels at long documents, nuanced reasoning, and creative tasks. Claude Haiku is available as a fast cheap option for simpler operations.

Cost: ~$0.01–0.05 per conversation

Security & Privacy

Your data never leaves
without your command.

🚫

Zero Telemetry

QuintX sends no usage data, analytics, or crash reports to any server ever. What happens on your PC stays on your PC.

🧠

RAM-Only Screenshots

Screen captures go to RAM then directly to your AI. Never written to disk. Never saved anywhere. Zero storage footprint.

🔑

Local Key Storage

API keys stored only in .env on your PC — never uploaded, never sent to QuintX. We have no servers that receive them.

✋

Explicit Permissions

Every tool requires you to tick each permission before activating. You always know exactly what QuintX can access.

⚠️

Delete Guard

Delete, erase, format, wipe — any of these in your message triggers a confirmation. Must type YES DELETE exactly. No accidents.

📴

Offline Capable

Ollama + Guide Mode = zero internet required. All data, all AI, all processing stays on your local machine.

Help & Troubleshooting

Common questions
answered.

Engine offline error

Check your .env has the correct AI_PROVIDER and API_KEY. If using Ollama, make sure it's running: open a terminal and run ollama serve. Then run ollama list to confirm your model is installed.

No module named X error

The auto-installer should handle this. If it fails, run: pip install -r requirements.txt from your project folder. If a specific package fails, install it manually: pip install package-name.

QuintX is running slowly

You're probably using Ollama with a large model on a slow PC. Switch to Groq (free, fastest) at console.groq.com. In Settings: set provider to OPENAI, paste your Groq key, set model to llama-3.1-70b-versatile.

Mic button not working

PyAudio is not installed. Run: pip install SpeechRecognition pyaudio. On Windows you may also need to install PortAudio. The mic button will turn green when audio is ready.

Tools panel is empty

The tools_config.json file is missing from your project root folder. It should be at C:\quintx-automation\tools_config.json. It's created automatically on first run — try restarting QuintX.

torchaudio DLL error

A conflicting package was installed. Run: pip uninstall torchaudio torch torchvision -y then restart QuintX. These packages are not needed and cause DLL conflicts on Python 3.13.

Sidebar in wrong position

QuintX uses the Windows API for screen detection. If your display scaling is not at 100% or 125%, the position may be off. Try changing Windows display scaling, or drag the panels manually to your preferred position.

Telegram bot not responding

Check the bot token is correct (from @BotFather). Make sure QuintX is running on your PC — the bot only works while QuintX is open. Check the status shows ● ONLINE in the Bots panel. Try stopping and starting again.

WhatsApp webhook not receiving

Make sure ngrok is running: ngrok http 5000. Copy the https URL and paste it into Twilio's webhook setting. The URL changes every ngrok restart — update it in Twilio each time unless you have a paid ngrok account.

How to add a custom tool

Create a Python file in tools\ with a function decorated with @tool from langchain_core. Add an entry to tools_config.json. Then add the import in core/engine.py under _load_tools().

QuintX_OS

Your AI. On yourscreen. Always.

Everything in one panel.

Web Search & Browse

File Management

Voice Commands

Image Analysis

WhatsApp & Telegram

Tools Manager

Built-in Guide LLM

Delete Guard

Screen Vision

Modules & Sub-modules

How QuintX processesyour requests

You type or speak

Destructive keyword check

AIThread starts in background

LangChain agent selects tools

Steps appear live in ResultPanel

Final answer as chat bubble

12 tools. Every onepermission-controlled.

Web Search

Open URL

Web Scraper

Read File

Write File

List Folder

Screen Vision

Email Agent

App Control

Code Runner

Image Gen

Custom Slot

The AI brain that picksthe right tool every time

User Input Received

LLM Reasons & Plans

Tool Called with Arguments

Observation Fed Back

Final Answer Displayed

Reply on WhatsApp &Telegram automatically

What you configure

Your AI. Your choice.

Your data never leaveswithout your command.

Zero Telemetry

RAM-Only Screenshots

Local Key Storage

Explicit Permissions

Delete Guard

Offline Capable

Common questionsanswered.

Built for the long game.

Core Automation

Vision & Control

Agentic OS

Download QuintX OS.Run in 60 seconds.

Your AI. On your
screen. Always.

How QuintX processes
your requests

12 tools. Every one
permission-controlled.

The AI brain that picks
the right tool every time

Reply on WhatsApp &
Telegram automatically

Your data never leaves
without your command.

Common questions
answered.

Download QuintX OS.
Run in 60 seconds.