01 // Before you start
What you need.
A Windows PC and a few minutes. Anything marked optional can be added later, whenever you like.
- RequiredA Windows 10 or 11 PC, 64-bitAny PC works. With an NVIDIA graphics card with 6 GB or more, the AI can run free on your PC and voice is fastest. Without one, use an AI account instead.
- RequiredAn internet connectionFor the licence check when it starts (an answer stays good for 7 days offline), for cloud AI if you use it, and for the weather, music and other services. Local-AI mode uses models on your PC; premium voices and connected services have separate settings.
- RequiredA brain for your assistantPick one or both. Free on this PC: it installs Ollama and downloads a model for you. Or your own account with Claude, OpenAI, Google Gemini or OpenRouter: paste a key and pay them only for what you use.
- OptionalA microphone and speakersTo talk to it. Everything works by typing too: press Enter on any screen.
- OptionalA webcamSo it notices when you come and go, and for hand gestures. Off unless you turn it on.
- OptionalYour other accountsSpotify, Google Calendar and Gmail, Discord, Steam, Home Assistant and Claude Code. Each is connected later, in the studio’s Connectors.
Will my PC run it?
Yes. The setup reads your graphics card and memory, picks the level that suits them and tells you why, in plain words. You can choose another in the setup’s Your PC step.
| Level | Typical PC | What you get |
|---|---|---|
| Lite | Graphics built into the processor, as on most thin laptops | The lighter, flat look. Cloud AI is the better choice. |
| Standard | A graphics card | The full 3D core on one screen, and small local models. |
| Multi-screen | Standard, plus more screens or devices | The same, spread across them. |
| Pro | 8 GB or more of graphics memory and 16 GB of memory | Everything, including the biggest local models. |
Graphics cards and memory
Graphics memory (VRAM) decides which free local model runs well. As a guide, from the same check the setup runs:
- 16 GB+Big local models, up to about 30B
- 10 GB+Local models up to about 14B
- 6 GB+Up to about 8B
- 4 GB+Small models, 3 to 4B
- LessCloud AI recommended; tiny local models only
NVIDIA cards are best: local AI is fastest, and listening can move onto the card after a one-time download of about 1 GB. AMD and Intel Arc cards are recognised and get the right level, but show no live load or temperature. With no graphics card at all, local AI still works, more slowly, and speech is understood on the processor.
Leave room for what you choose: the installer is about 120 MB, Ollama about 1.7 GB, a free model 1.4 to 9.3 GB, and faster NVIDIA voice about 1 GB. The Doctor warns you when less than 2 GB is free.
02 // Installing
Install it.
One file, installed just for you. No administrator password and nothing to configure.
Check the download (optional)
In PowerShell, this prints the file’s fingerprint:
Get-FileHash "$HOME\Downloads\HoloIntelligenceSetup.exe" -Algorithm SHA256It should match: REPLACE-AT-RELEASE
The fingerprint for each release is listed on the download page.
-
Run HoloIntelligenceSetup.exe
Open it from your Downloads folder. It installs just for you, so Windows won’t ask for an administrator password.
-
If Windows SmartScreen asks
Until the installer is code-signed, SmartScreen may stop it with a blue “Windows protected your PC” box. Choose More info, then Run anyway.
Illustration: the two clicks. SmartScreen warns about any app that isn’t code-signed yet, which is why it shows “Unknown publisher”. -
The setup opens by itself
After a few seconds the setup wizard opens. If you close it, open it again from the desktop shortcut or the Start menu (search for Holo).
-
It starts with Windows
From then on its parts start when you sign in, so your assistant is ready when you are. To hide or show its screens, press Ctrl+Alt+H.
Your free trial and your key
Every PC gets a free 14-day trial. When you buy, your key arrives by email. Open the studio (press T), go to Account, paste it and choose Use this key. Keys look like HOLO-XXXXX-XXXXX-XXXXX-XXXXX.
- One key works on up to 3 PCs and includes a year of updates. The Account tab shows the date your updates run to.
- It checks the licence when it starts and once a day. An answer stays good for 7 days without internet.
- When a trial ends, the screens show a card where you can paste a key or buy one. The studio stays open.
- Moving to a new PC? Choose Remove the key from this PC in Account first, and its place is free for another.
03 // The setup wizard
The setup, step by step.
Ten short steps, about three minutes. Each one saves as you go, so closing it halfway loses nothing, and all of it can be changed later in the studio.
At the top of every stepFull guide opens this guide inside the app. Skip setup goes straight to the home screen.
-
Step 01 of 10
Welcome
Your assistant’s body sits beside every step, and reacts to what it hears, thinks and says. The first screen lists what’s coming: a name, a personality, a look, a voice, an AI, a check of your PC, and what it may see and hear.
-
Step 02 of 10
What you’ll need
A quick checklist so nothing is a surprise: the same list as Before you start. Nothing on it has to be ready now.
-
Step 03 of 10
Name
Pick any name: Nova, Jarvis, your cat’s name. It’s also the wake word, so short names of one or two syllables are easiest to say and to recognise. Then choose what it calls you: sir, boss, captain, your name, or nothing.
-
Step 04 of 10
Personality
Butler (dry British wit), Friendly, Minimal or Playful, then three sliders: humour, formality, and how often it speaks up on its own. Personality changes how it talks, not what it can do.
-
Step 05 of 10
Look
Choose a theme (Ember, Arctic, Amber, Violet, Emerald, Graphite or Daylight) and a body (Core, Wave, Swarm, Lattice or Face). Comfort settings live here too: high contrast, calm motion, a readable font, captions and text size.
-
Step 06 of 10
Voice
Eight free voices that run on your PC, British and American: tap one to hear it. Set the speed, or add an ElevenLabs key for premium voices. With an NVIDIA card it offers a one-time download (about 1 GB) that makes listening several times faster.
-
Step 07 of 10
Connect an AI
The two options side by side: install Ollama with one click for free local AI, or paste a key from Claude, OpenAI, Google Gemini or OpenRouter and press Check and save. Use either, or both. Giving it a brain covers each in detail.
-
Step 08 of 10
Your PC
It reads your graphics card and memory, picks a level (Lite, Standard, Multi-screen or Pro) and says why. Lite uses less power and suits any laptop. You can override it here, and it reminds you that phones, tablets and other computers can join as extra screens later.
-
Step 09 of 10
Privacy
Microphone, camera, clipboard history, crash reports and usage statistics: each one is off until you switch it on. Off means off: the camera light goes out and the microphone isn’t opened.
-
Step 10 of 10
Ready
A summary of your choices, then Open my home screen. The first time, seven short cards show what’s where and which keys do what. Press ? to see them again.
04 // Giving it a brain
Give it a brain.
Your assistant needs an AI model to think with. Use the free one on your PC, your own account, or both: quick things to the free model, hard ones to your account.
You choose which model does what in the studio’s Models tab: chatting, coding, seeing pictures, voice and building. Each choice is described by what it’s like, such as “fast and free, stays on this PC”.
Option 1: free, on your PC (Ollama)
- In the setup’s Connect an AI step, press Install Ollama for me. It downloads Ollama (about 1.7 GB) from ollama.com, checks the download is genuinely signed by Ollama Inc., and installs it just for you, with no administrator password.
- Press Download to get the free model that suits your graphics card.
- That’s it: nothing you say leaves your PC. With no NVIDIA card it still works, more slowly.
Already have Ollama? Choose I’ve installed it myself: check again. Models running in LM Studio on this PC are found too. You can do all of this later from the studio’s Models tab or the Doctor.
| Graphics memory | Model it picks | Download |
|---|---|---|
| 10 GB or more | Qwen3 14B | about 9.3 GB |
| 6 GB or more | Qwen3 8B | about 5.2 GB |
| 3.5 GB or more | Qwen3 4B | about 2.6 GB |
| Less, or none | Qwen3 1.7B | about 1.4 GB |
The biggest model that fits your card comfortably. Without a graphics card, the 1.7B model is small enough to run on the processor.
Option 2: your own AI account (an API key)
An API key is a password that lets Holo Intelligence UI use your account with an AI company. You paste it once. It’s checked with the company first, then kept encrypted on your PC only, and never shown again.
You pay the AI company directly. Costs depend on your model, the amount of text and the tools you use. Check the provider’s current prices before adding credit. The daily limit in the studio’s Models tab switches answers to a local model when reached; it is not a billing cap on your provider account.
sk-ant-Claude (Anthropic)
The smartest choice for building new features, and excellent for everything else.
- Open console.anthropic.com and sign in, or create an account.
- Add credit under Plans & Billing. You pay as you go, from $5.
- Go to Settings, then API keys, and choose Create Key. Name it “Holo”.
- Copy the key (it starts with
sk-ant-) and paste it into the setup. It’s shown only once.
A Claude.ai Pro or Max subscription is separate from API credit. To use your Claude plan for building, connect Claude Code instead (Connectors).
sk-OpenAI
GPT models: good all-rounders, and the same key works for premium voices.
- Open platform.openai.com and sign in.
- Add a payment method or credit under Settings, then Billing.
- Open API keys and choose Create new secret key. Name it “Holo”.
- Copy the key (it starts with
sk-) and paste it into the setup.
ChatGPT Plus doesn’t include API use: the API is billed separately, per use.
AIzaGoogle Gemini
Gemini models, very good at seeing pictures. There’s a free tier with daily limits.
- Open aistudio.google.com and sign in with your Google account.
- Choose Get API key, then Create API key.
- Copy the key (it starts with
AIza) and paste it into the setup.
The free tier is enough to try it. Add billing in Google Cloud for more.
sk-or-OpenRouter
One key for hundreds of models from many companies, pay as you go.
- Open openrouter.ai and sign in.
- Add credit under Credits.
- Open Keys and choose Create Key.
- Copy the key (it starts with
sk-or-) and paste it into the setup.
Some models on OpenRouter are free, with lower limits.
Then press Check and save. It tries the key, tells you how many models it found, and keeps it encrypted on this PC. You can add or replace keys any time in the studio’s Connectors.
Which models can build?
Before a model may build a plugin, it takes a short intelligence check: five coding tasks. Four out of five earns can build plugins; two earns can customize (themes, layouts, panel settings and routines); fewer is chat only. The badge shows in the Models tab. Check the preview before installing anything a model builds.
Local-AI mode (Models) lets only models on your PC answer, build and see. Cloud builders such as Claude Code are refused while it’s on. Premium voices can still send reply text to their provider, and connected services still contact their services. Choose a local voice and disconnect services you do not want to use.
05 // Connecting everything else
Connect the rest.
Music, your calendar, games, your smart home and your coding agents. All optional, all added in the studio whenever you like.
- Premium voicesStudio · Assistant
ElevenLabs (elevenlabs.io, Profile, then API keys; a free monthly allowance) or the same OpenAI key as above. Your free voices always work.
- Faster voice on NVIDIA cardsSetup · Voice, or Studio · Assistant
A one-time download of NVIDIA’s speech libraries (about 1 GB), so speech is understood several times faster.
- Claude CodeStudio · Connectors
Install it from claude.com/claude-code and sign in with your Claude plan. Connect it to see every session live and answer its permission requests from any screen. It can also build new panels for you.
- Codex CLI and Hermes AgentAutomatic
Their sessions appear beside Claude Code’s on the Claude Code and Agents panels. Permission requests come only from Claude Code.
- SpotifyStudio · Connectors
Music on every screen and by voice (“play some jazz”). Controlling playback needs Spotify Premium.
- Google Calendar and GmailStudio · Connectors
Your day on the screens and reminders before meetings, through your own Google sign-in, set up once on the connectors page.
- Discord and SteamStudio · Connectors
Who’s online and in voice; your games and friends. Discord needs a bot token, Steam a free Web API key from steamcommunity.com/dev/apikey.
- Home AssistantStudio · Connectors
Lights, plugs, heating and scenes by voice. In Home Assistant: your profile, Security, then Long-lived access tokens, Create token. Paste it with your Home Assistant address.
- Phones and tabletsStudio · Screens
Turn on “Let phones and tablets on my Wi-Fi join”, then scan the code. See More screens.
- Holo Link SoonStudio · Link
Connect with friends who have Holo Intelligence UI: notes, questions for each other’s assistants, and files, end-to-end encrypted. It needs Holo Cloud, which isn’t live yet.
06 // Using it
Talk to it.
Say its name, then what you want, or press Enter on any screen and type. Ask it anything, or ask it to change itself.
The microphone is off until you switch it on, in the setup’s privacy step or the studio’s Security tab. Typing always works.
Things to say
Tap one to copy it.
Keyboard shortcuts
| Keys | What it does |
|---|---|
| Enter | Type to your assistant |
| T | Open the studio |
| E | Arrange a screen: move, resize, add and remove panels, then Save |
| Shift+E | On the 3D screen: choose what each glass panel shows |
| S | Scenes, including Focus |
| A | The AI panels on the 3D screen |
| V | Choose speakers or enable push-to-talk |
| Ctrl+Alt+Z | Push-to-talk, once enabled in the voice menu |
| ? | The tour again |
| Ctrl+Shift+Z | Undo the last change |
| Ctrl+Alt+Z | Push-to-talk, once you switch it on (press V). Works anywhere in Windows. |
| Ctrl+Alt+H | Hide or show the screens. Works anywhere in Windows. |
| Esc | Close a menu or the tour |
The studio
Press T on any screen. Everything the setup set, and the rest, lives in its tabs:
- Look
- Assistant
- Models
- Connectors
- Build
- Plugins
- Gallery Soon
- Routines
- Crews
- Meetings
- Screens
- Link Soon
- Memory
- Security
- History
- Doctor
- Account
Want something it can’t do yet? Ask for it. The builder writes a panel, theme or routine, shows you a live preview, then installs it on the screen you pick, with one-click rollback. How plugins work.
07 // More screens
Add more screens.
It starts on one screen. Any phone, tablet or other computer on your Wi-Fi can join as another screen, in its browser. There’s no app to install.
Illustration: the studio’s Screens tab, and a phone showing approvals, Claude Code and notifications.
Open Screens
Open the studio (press T) and go to Screens.
Let them join
Switch on Let phones and tablets on my Wi-Fi join. The hub restarts for a few seconds. If Windows asks about its firewall, allow Private networks.
Scan the code
Point the phone’s camera at the code that appears, and open the link.
Keep it handy
Add it to the home screen to use it like an app.
A tablet or another computer joins the same way. Joined screens use the lighter, flat look, and show approvals, Claude Code, notifications or whatever you put on them.
What each screen shows
In Screens, pick what each screen shows: Its own panels, Mixed, AI command (models and agents) or My arrangement. Press E on a screen to move, resize, add and remove panels by hand, then Save. Press S for scenes, which rearrange every screen at once.
The code carries your screens’ key, so anyone on your Wi-Fi who has it can open a screen. Switch joining off again and only this PC can open screens.
08 // Privacy and safety
Off until you say so.
Local first. Nothing listens, watches or reads your clipboard unless you switch it on, and every change can be taken back.
- Microphone, camera and clipboard history: offEach stays off until you turn it on, in the setup or the studio’s Security tab. Crash reports and usage statistics are off unless you turn them on too.
- Keys encrypted on your PCEncrypted by Windows and readable only on this PC. Each key is checked before it’s kept, and never shown again.
- Local-AI modeOnly models on your PC answer, build and see, so every word stays on your PC.
- Only this PC opens screensUntil you let phones and tablets on your Wi-Fi join, in Screens.
- “What’s on my screen?” only when askedThe screenshot goes only to the vision model you chose, and is never stored.
- Meeting notes only when you start themOnly the words are kept, encrypted on your PC; audio never is. Recording a conversation can need everyone’s consent where you live, so let people know.
- Plugins in a sandboxNo network, and only the live data their manifest lists. Never your settings or keys.
- Claude Code, connected carefullyConnecting adds one approval hook to its settings, keeps the hooks you already have, and writes a backup first.
- Undo anythingCtrl+Shift+Z takes back the newest change. Safe mode turns every plugin off if something misbehaves (studio, History).
- An audit log in plain wordsEverything the assistant and its agents did, in the studio’s Security tab.
- Kid-safe modeAge-appropriate answers and risky tools off, with a PIN to switch it off again (studio, Assistant).
This website sets no cookies and runs no analytics. Read the privacy policy.
09 // If something’s wrong
Ask the Doctor first.
The studio’s Doctor tab checks every part in plain words, problems first, and fixes most of them with one button.
- ✕The assistantIts brain isn’t running, so it can’t answer.
- !Free local AIOllama isn’t installed. It’s free and runs AI on this PC without an account.
- !Voice speedListening runs on the processor. Your NVIDIA card could do it several times faster (about 1 GB, once).
- ✓ComputersThis PC is connected.
- ✓Screens1 open.
- ✓Your keysEncrypted by Windows, readable only on this PC.
Open it from the studio: press T, then Doctor. Its one-button fixes start a part that isn’t running, open the screens, leave safe mode, encrypt keys, install Ollama and download the faster NVIDIA voice. Anything else points you to the right studio tab. The same tab can also benchmark this PC: it times each local model and says whether it’s quick enough to talk to.
Common problems
Q.01Windows won’t run the installer
That’s SmartScreen, which warns about apps that aren’t code-signed yet. Choose More info, then Run anyway. See Installing.
Q.02It doesn’t answer
Open the Models tab. Is a model connected? Is local-AI mode on with no model on this PC? Has the daily limit moved it to the free model? The Doctor’s AI models row says which.
Q.03It can’t hear me
Check the microphone switch in the studio’s Security tab, and that a microphone is plugged in or chosen in Windows’ sound settings. Say its name first, then what you want. Typing always works: press Enter.
Q.04It speaks from the wrong speaker
Press V on a screen to choose where it speaks.
Q.05The screens have disappeared
Press Ctrl+Alt+H to show them, or open it from the Start menu (search for Holo). The Doctor’s Screens row has an Open it button.
Q.06Something looks wrong after a change
Ctrl+Shift+Z undoes the newest change; press it again to go further back. The studio’s History tab lists every change. Safe mode turns plugins off and restores the default look. A screen that fails to start offers Undo the last change, Start in safe mode or Reload, and a 3D screen falls back to the flat look first.
Q.07The 3D feels slow
Choose the Lite level in the setup’s Your PC step for the lighter, flat look, or switch on Calm motion in the studio’s Look tab: no drifting, breathing or sliding.
Q.08My phone can’t connect
Check the phone is on the same Wi-Fi, that joining is switched on in Screens, and that Windows’ firewall allows it on Private networks. Scan the code again after switching joining on, since the hub restarts.
Q.09Claude Code’s approvals don’t appear
Connect Claude Code in the studio’s Connectors. If nobody answers within about 108 seconds, or the hub is down, Claude Code simply asks in its own window as usual. Only Claude Code’s requests come to the screens.
Q.10An API key won’t save
Check you copied the whole key and that it starts with the right letters (sk-ant-, sk-, AIza or sk-or-), and that your account has credit or billing set up. A ChatGPT Plus or Claude.ai subscription doesn’t include API use.
Q.11It says my trial has ended
Paste your key in the studio’s Account tab, which stays open; or buy one from the card on the screens. It needs the internet at least once every 7 days to refresh its licence.
Still stuck? Help and answers Contact us
10 // Updating and uninstalling
Update it. Or remove it.
Both take a minute, and neither touches your settings unless you say so.
Run the newer installer
Download the newest installer and run it. It stops the running copy, swaps in the new files, and keeps your settings, keys, memories and plugins.
There’s no automatic updater yet, so check the download page now and then. A key includes a year of updates: the Account tab shows until when.
Remove it from Windows
- Optional: choose Remove the key from this PC in the studio’s Account tab, so its place is free for another PC.
- Open Settings, then Apps, then Installed apps (Apps & features on Windows 10). Find it (search for Holo) and choose Uninstall.
- It asks whether to delete your settings, keys and what the assistant remembers. No, the default, keeps them for a later reinstall.
- Ollama is a separate app. Uninstall it the same way if you no longer want it.
That’s the whole guide
Install it, name it, give it a brain. Then just talk to it.