Listen
Qwen3-ASR
Your voice becomes text on your own machine. The language is detected as you speak.
Computer-use agent for Windows
Speak.
Apex does the clicking.
Apex sits on the right edge of your desktop, listens, and operates the computer for you — apps, windows, the browser, files and documents. On your own machine, with models that run locally.
Download now
ApexHere.msi · Windows 10 / 11, 64-bit
- 30+
- languages it hears and speaks
- 90+
- tools for screen, mouse, browser and files
- 0
- accounts needed
01 In action
One sentence.
Every step on the record.
Apex shows each action as it happens — what it opened, where it clicked, what it read. Anything that changes your files waits for your yes.
19°
02 Capabilities
Everything you do with a mouse and keyboard.
Said once.
-
01
Voice
Push-to-talk or continuous conversation. Interrupt it mid-sentence and it stops to listen. Speech recognition and a natural voice in 30+ languages.
-
02
Eyes
Reads the screen with Windows OCR and UI Automation, so it finds buttons by name, not by guesswork. With a vision model it sees images too.
-
03
Hands
A real mouse and keyboard. Launches apps, focuses, moves, minimises and closes windows, types into whatever has focus.
-
04
Browser
Its own Chromium with sign-ins that stay. Opens pages, reads them, fills in forms and clicks the way a person does.
-
05
Files
Searches the Windows index, opens folders in Explorer, copies and moves. Deleting goes to the Recycle Bin — never further.
-
06
Documents
Writes Word, PDF and Excel files in any font installed on your PC. Screenshots it takes are saved next to them.
-
07
Memory
Remembers what you tell it, keeps the steps of tasks it has solved, and searches earlier conversations. All of it on your disk.
-
08
Scripts
Runs PowerShell or cmd when you allow it — with a time limit, and a key that stops it instantly.
03 How it works
Listen. Think. Act.
Then say what it did.
Think
Qwen3.5-4B · vision
A local model plans the steps and picks the tools. Prefer the cloud? Connect OpenAI, Anthropic, Google or Ollama instead.
Act
Supertonic 3
Tools do the work, one visible step at a time. The answer is spoken sentence by sentence, while it is still being written.
04 Settings
Every switch
within reach.
Models, voice, hotkeys, memory, permissions, MCP — one window. No config files, no command line.
General
Türkçe
33%
Open the panel when ApexHere starts
Hide the panel from screen captures
Run ApexHere as administrator
Agent
25
Speech
Qwen3-ASR 0.6B
M1M2F1F2F3
Automatic · GPU
Speak replies out loud
Let me interrupt Apex by speaking
Hotkeys
CtrlAltSpace
CtrlAltEsc
Models
Qwen3.5-4B · Q4_K_M2.74 GBInstalled
Qwen3.5-4B · vision (mmproj)0.68 GBInstalled
Qwen3-ASR 0.6B1.15 GB
Supertonic 30.45 GBInstalled
Active language model
Local · Qwen3.5-4Bon this PC · vision
OpenAIcloud
Anthropiccloud
Googlecloud
Ollamayour server
About you (USER.md)
- Prefers short answers. - Invoices are in Documents\Finance. - Uses Chrome; signed in to the bank portal.
Skills
send-whatsapp-messageopen WhatsApp Web, find the contact, type, send
monthly-invoice-summarysearch *.pdf in Finance, read totals, write a PDF
Permissions by risk
Readauto
Interactauto
Modifyask
Destructiveask
Executeask
Allowed folders
C:\Users\you
D:\Projects
ApexHere MCP server
Serve tools over MCP
http://127.0.0.1:47900/mcp
••••••••••••••••
External MCP servers
filesystemstdio · connected
crm-apihttp · connected
05 Runs locally
Runs here.
Not in someone else's cloud.
With the local models, your voice, your screen and your files never leave the PC. A cloud model is there when you want one — your setting, not ours.
- Models on disk
- ~5 GB
- Accounts required
- 0
- Hardware
- GPU / CPU
Speech, language and voice, downloaded once from Settings.
No sign-up to use it. Install and talk.
An NVIDIA GPU makes it quick. Without one, it all runs on the CPU — or a cloud model does the thinking.
06 Control
It acts.
You keep the keys.
- Risk levels
Reading happens on its own. Changing files, deleting and running scripts ask first — and you decide per tool.
- Yes, by voice
Approve or refuse out loud. “Yes” and “no” are enough.
- One stop key
A single hotkey stops speech, tools, scripts and any key it is holding down.
- Allowed folders
File tools work only inside the folders you list.
- Administrator, if you say so
Off by default. Turn it on and Apex can also work in apps that run elevated.
07 MCP
Plugs into the rest
of your AI tools.
← in
An MCP server
Apex offers its tools over the Model Context Protocol. Claude Desktop, Cursor or your own agents can use this PC through it — with the same permissions.
out →
An MCP client
Connect other MCP servers and Apex borrows their tools: your CRM, your database, your internal services.
08 Specifications
- Platform
- Windows 10 (2004 or later) and Windows 11, 64-bit
- Install
- One MSI. Self-contained — nothing else to install.
- Hardware
- NVIDIA GPU recommended · CPU supported
- Speech
- Qwen3-ASR · Supertonic 3
- Local model
- Qwen3.5-4B with vision
- Cloud models
- OpenAI · Anthropic · Google · Ollama · OpenAI-compatible
- Languages
- 30+ for speech · app interface in English
- MCP
- Server and client
Put Apex
on your desktop.
Download now
ApexHere.msi · Windows 10 / 11, 64-bit
09 Questions
Does it need the internet?
Only to download the models once. With the local models it works offline — except for the browser tool, which of course needs the web.
What happens to my screen and voice?
With the local models they are processed on your PC and go nowhere else. If you choose a cloud model, what you say and what Apex sees is sent to that provider.
Can it do something I didn't want?
Every action is shown as it happens. Changing or deleting files and running scripts ask for approval by default, file tools stay inside your allowed folders, and the stop key ends everything at once.
Where are my files and its memory kept?
Documents and screenshots go to Documents\ApexHere. Its memory — notes, saved skills and earlier conversations — stays in its install folder. You can see and edit all of it in Settings.
Which languages does it speak?
It understands and speaks 30+ languages, including Turkish and English. The app's own interface is in English.
I don't have an NVIDIA graphics card.
Apex still runs. Speech recognition and the voice switch to the CPU, and you pick a cloud model for the thinking in Settings.