Features
Built to run on your terms
A desktop app that brings open AI models onto your computer and gives them the web and a browser. You stay in charge of both.
Version 0.1.7 · Windows x64 · macOS Apple Silicon · Free
Runs on your computer
The right model for your machine, every time
X Alpha AI checks your memory, graphics card and disk, then shows which models will run well, which will be slow, and which won’t fit, before you download anything.
- “Best for this computer” picks a strong model that runs comfortably
- Downloads pause and resume, and are checked when they finish
- Start a model and chat offline. Nothing leaves your computer
- Context window sized automatically to your memory
Best for this computer
- QRuns wellStart
Qwen3 8B
Qwen (Alibaba) · 5.0 GB
- GRuns well64% · Pause
Gemma 3 12B
Google · 7.3 GB
- MRuns slowlyDownload
Mistral Small 24B
Mistral AI · 14.3 GB
- MToo big
Llama 3.3 70B
Meta · 42.5 GB
Searches the web
Fresh answers, with sources you can check
Turn on Web or type /web and the app searches, reads the pages, and answers with numbered sources. Built-in search works out of the box; your own SearXNG server is optional.
- Clickable sources under every web answer
- Paste a link and the page is read for you
- /pkg gives exact latest versions from npm, PyPI, crates.io and GitHub
- When the AI wants to look something up itself, it asks first
Ask anything…
Running on this computer · works offline
Uses a browser for you
It browses for you. Only where you say yes.
Ask it to open a site, read a page that needs JavaScript, click through or fill something in. The first time it wants to act on a site, you choose Allow or Deny.
- Reads pages built with JavaScript that plain search can’t
- Clicks, types and scrolls while you watch every step
- Permission per site, and you can stop at any time
- Works on Windows and macOS
Use the browser on portfolio.dev?
The AI will read the page and may click or type there.
Reads your files and images
Drop in a file. Ask anything about it.
Add files with the + button or drag them in. Models that can see read images directly; documents are read as text.
- PDF, Word, text and code files
- Images read by vision models, fast even on a laptop
- Everything is processed on your computer
Do the totals on these two match?
Not quite. The invoice total is ₹18,640; the receipt shows ₹18,460. The digits 4 and 6 are swapped. Everything else (date, GST number, items) matches.
Ask about your files…
Private by design
Your chats. Your models. Your keys.
Models run on your own machine on 127.0.0.1. Chats are saved locally. Network requests go through a guard that refuses private and local addresses.
- Sign in once; local models then work offline, and chats never leave this computer
- API keys (when you add cloud models) live in your OS keychain
- The AI asks before each lookup or browser action
Your computer
Models
Run on 127.0.0.1
Chats
Saved on this computer
API keys
In your OS keychain
Only web searches and pages you allow leave the computer, through a guard that blocks private and local addresses.
The details
Small touches you’ll notice every day
Saved chats
Every chat is kept on your computer. Rename or delete from the sidebar; they reopen after a restart.
Stop any time
Stop an answer mid-sentence. Code blocks have Copy; answers are rendered as clean Markdown.
Model settings
Context window (automatic or your choice) and temperature, with the memory each size needs.
Downloads that survive
Pause and resume, even after a restart. Every download is checked when it finishes.
Reuses your models
Models already downloaded by our LocalPilot VS Code extension are used in place, not downloaded again.
Light and dark
Follows your system theme.
Real speeds
Each model is measured on its first start, so speed estimates match your machine.
Slash commands
/web to search, /pkg for exact latest versions from npm, PyPI, crates.io and GitHub releases.
Windows and macOS
NVIDIA (CUDA) and AMD (Vulkan) graphics on Windows, or the processor alone; Metal on Apple Silicon.
Coming soon
What’s coming next
Built in steps. Each one ships when it works well on real computers.
Your own cloud keys
Use OpenAI, Anthropic, Gemini or OpenRouter models with your own key, alongside local ones.
Spaces and memory
Separate Personal and Work spaces; memory you can see, edit and delete.
Images and video
Generate images (cloud or local) and short videos, with a cost estimate first.
Skills
Install skills that teach the assistant new jobs, with checks that show they work.
Teams
One subscription per team, shared keys through a relay, admin rules and usage per member.
Your AI is ready when you are
Free to start with a free account. Local models, web search and browser control, offline after you sign in once.
Version 0.1.7 · Windows x64 · macOS Apple Silicon · Free