SMARAN.AI Documentation
How to install SMARAN.AI, what each part does, and where your data is kept - on Windows, Linux and Android.
On this page
01 Quickstart Overview
SMARAN.AI is a personal assistant that runs on your own computer. It can run open models locally through Ollama, so a conversation with a local model never leaves your machine. When you want a bigger model, you can add a key for a cloud provider - then that request goes to that provider.
Your chats, documents, memory and API keys are stored in your own user folder. Nothing you type is sent anywhere unless you pick a cloud model, switch on web search, or use the Gemini voice engine.
🖥️ Desktop app
Windows installer or portable zip, and Linux AppImage, .deb, .rpm or .tar.gz. Python is bundled; local models need Ollama, which the app can install for you.
📱 Android app
Works on its own with a free Gemini, Groq or OpenRouter key, and links to your PC for local models, your documents and PC control.
⚡ VS Code extension
SMARAN.AI Codex: ask about the open file, propose edits and run commands, with a local model or your own cloud key.
02 Hardware Requirements
The app itself runs on any 64-bit Windows 10/11 PC, or Linux x86-64 with glibc 2.28 or newer. What you need beyond that depends on which model you run. Every model in the Model Hub shows its memory requirement before you download it: across the 63 models in the catalogue it runs from 4 GB to 512 GB, and 18 of them need 8 GB or less.
- A GPU makes local models faster; it is not required.
- Cloud models need no particular hardware - only a key and a connection.
- A webcam only if you want vision from the camera.
03 Installation Steps
Download from the download section. Step-by-step instructions for each platform, including the Windows SmartScreen warning (the app is not code-signed yet), are in the install guide.
chmod +x SMARAN.AI-x86_64.AppImage
./SMARAN.AI-x86_64.AppImagecurl.exe -L -o smaran.exe https://github.com/SHASHWAT-MISHRA-997/SMARAN.AI-downloads/releases/latest/download/smaran.exe04 Model Hub
The Model Hub lists 63 models, filtered by what they are good at - reasoning, vision, code, text, audio, video. It keeps three things apart: a model that is only in the catalogue, one you have downloaded, and one that is actually running.
Local models
Downloaded through Ollama and kept on your drive. They cost nothing to run and work offline.
Cloud models
Add a key for a provider such as Gemini, Groq or OpenRouter. The key is checked before it is saved, and the provider bills you directly, if at all.
Pictures and video
On this PC, or on the free cloud: FLUX (free NVIDIA key) or Pollinations for pictures, Wan 2.2 or LTX-Video on Hugging Face's free shared GPUs for short video, which has a daily allowance.
05 SMARAN Code
SMARAN Code works on a project folder: it reads the files, edits them, runs commands and tests, and fixes what fails. Approvals are Manual, Smart or Off, and a whole run can be undone.
Commands run with filters and time limits, and files are checkpointed before a run so they can be restored. This is not an operating-system sandbox: a command you approve runs as you, with your permissions.
06 Voice & Speech
The live voice call has two engines:
- Local: faster-whisper hears you, a local model answers, Kokoro speaks - all on this machine, with the network off. Kokoro speaks English; other languages use an online voice.
- Reading a reply aloud (Speak): English is spoken by Kokoro on this machine. Other languages use Microsoft's online voice, so for those the reply's text is sent to Microsoft.
- Gemini: Google's live voice, which can also see your screen or camera. It needs a Gemini key, and your audio goes to Google while the call is open.
The wake word: on the phone "Hey SMARAN", "Hey Jarvis", "Hey Amarya" and "Hey Myra" are all heard offline; on the PC "Hey SMARAN" and "Hey Jarvis" are heard offline.
07 Design Studio & Sites
Describe a page and SMARAN writes it as HTML, CSS and JavaScript, with a live preview beside the prompt. It runs on whichever model you have chosen - local or cloud.
🎨 Design Studio
Generate a page or component from a sentence and keep refining it in the same conversation.
🌐 Sites
Your designs are kept, so you can come back to one, keep refining it, or download it as an HTML file.
🔒 Separate history
Design work is kept apart from your regular chats.
Choose Design Studio (under Sites) in the sidebar of the desktop app, and type what you want - for example "a dark pricing page".
08 VS Code Extension
SMARAN.AI Codex works beside your code in VS Code, Cursor or Windsurf. It talks to a local model server (Ollama or LM Studio) or to a cloud provider with your own key. Files it reads are sent to the model you chose.
💬 Ask and edit
Explain or refactor the open file, write tests for it, or fix the error under the cursor.
🛡️ You choose its reach
Read-only, workspace or full access, and approval for every action, only commands, only risky ones, or none.
🎙️ Dictation
Dictate a prompt instead of typing it.
Search "SMARAN.AI Codex" in the Extensions view, or run ext install ShashwatMishra.smaran-ai-codex from the command palette (Ctrl+P).
{
"smaran.provider": "",
"smaran.model": "",
"smaran.ollamaUrl": "http://localhost:11434",
"smaran.reach": "workspace",
"smaran.approval": "always"
}09 Android App & Linking Your PC
The phone app works on its own. Linking it to your PC adds shared history, your PC's local models and documents, and control of the PC from the phone:
- On the PC, open Settings and choose Link your phone. A QR code appears.
- On the phone, choose Link a computer and scan it - or type the pairing code shown under it.
- Both devices now use one account and one history. Unlink from either side at any time.
Only a fixed set of actions can be asked of the PC from the phone, and anything about money, buying or the power button still asks you first.
10 Where Your Data Lives
Everything the app keeps is in one folder. Deleting it deletes your chats, memory, keys and settings.
| System | Folder |
|---|---|
| Windows | %LOCALAPPDATA%\SMARAN.AI\data |
| Linux | ~/.local/share/SMARAN.AI/data (or $XDG_DATA_HOME/SMARAN.AI/data) |
| Android | The app's own private storage; removed when the app is uninstalled. |
Downloaded local models are kept by Ollama in its own folder.
11 Privacy & What Is Sent
- Cloud models: when you choose one, your message and the context it needs go to that provider.
- Web search: only when you switch it on.
- Reading a reply aloud: English stays here (Kokoro); other languages go to Microsoft's voice service.
- Gemini voice: your microphone audio (and screen or camera, if you turn them on) goes to Google while the call is open.
- Usage data: none. The app sends no install id, no launch or usage counts and no reports of any kind.
- This website: counts visits and download clicks (an event name only, no personal data), plus anything you type into the feedback form.
See also the Terms of Service and the privacy summary.