Via

Via answers one question: is this block getting better or worse over time? You mark the two ends of a city block in Google Street View. Via then looks at the oldest and newest public street imagery at several points along that block and, using a vision model that runs on your own computer, notes for each building frontage whether it has visibly improved, stayed the same, or declined. It gathers those notes into a one-page report about the block's direction.

It is for anyone choosing where to live, whether buying or renting, and anyone weighing an investment in a block. A visit shows you what a street is like today. Via shows you which way it has been heading.

Where Via comes from

Via, the Latin and Italian word for street, extends the "Compounding the Little Things" research behind Kai, the repository housing the Shrine Change Tracker by the same author. That project tracked small changes on shrine walls along the streets of Rome, streets literally named Via, on the thesis that small changes in public spaces go untracked until their cumulative effects are impossible to ignore. Via applies the same approach, public Street View history read by a vision model on your own computer, to the buildings people live in. The full story is in docs/lineage.md.

What you get

A report page you can read at a glance: the block's direction (improving, stable, or declining), the years compared, and how much of the block could actually be assessed. Below that, one card per sample point with the old and new views side by side, the verdict, and the specific changes observed, like "facade repainted" or "ground floor storefront now boarded". One button saves the whole thing as a report file you can keep or share.

Is this safe and private?

Yes. Here is exactly what each part does and does not do:

• Ollama is a free, open program that runs the vision model on your computer. It does not send your data anywhere and only answers when Via asks it a question.
• The local program (a small Python server) only talks to three things: Google Street View, to fetch public street images; a free public map service, to look up the shape of the street you marked; and Ollama, to look at the images. It cannot touch your documents, photos, email, or anything else.
• The Chrome extension only reads the web address of your Street View tab to get the location. It cannot see your browsing history, passwords, or other tabs.

No accounts and no keys are needed, and nothing about you or your computer is sent anywhere. Asking for street images and for the shape of a street does mean asking those two public services about the spot you marked, the same way opening a map in your browser does. Nothing else goes out, and the pictures and the findings stay on your machine. This tool does not save or collect Street View images. It saves only its own written findings. Images fetched during a run live in a temporary folder your system cleans out, and you can empty it yourself anytime with the "Clear cached images" button on the analysis page.

What this is not

• Not price prediction and not advice. This tool shows how buildings on a block have visibly changed over time. It does not predict prices and it is not financial advice.
• No claims about neighborhoods or people. Via reports visible building condition only. It says nothing about safety, crime, desirability, or who lives somewhere, and it never should.

Setup

You will do five things, once: install Ollama, download the vision model, download this project, set up the local program, and load the extension in Chrome. Copy and paste each command exactly.

What you need

• Google Chrome
• Python 3.9 or newer (check with python3 --version; if missing, install it from https://www.python.org/downloads/ and on Windows tick "Add Python to PATH")
• About 4 GB of free disk space for the vision model
• 8 GB or more of memory

Step 1: Install Ollama

Ollama runs the vision model on your computer.

Windows: download the installer from https://ollama.com/download/windows and run it. Ollama starts automatically and puts an icon in your system tray.

Mac: download from https://ollama.com/download/mac, open the file, and drag Ollama into Applications. Open it once; it appears in your menu bar.

Linux: open a terminal (usually Ctrl+Alt+T) and run:

curl -fsSL https://ollama.ai/install.sh | sh
ollama serve

Keep that terminal open while you use Via.

Step 2: Download the vision model

Open a terminal. On Windows, press the Start key, type cmd, and press Enter. On a Mac, open Terminal from Applications, then Utilities. Run:

ollama pull gemma3:4b

This downloads about 3 GB, once. Check it worked with ollama list.

Step 3: Download this project

Click the green Code button at the top of this page, choose Download ZIP, and unzip it somewhere easy to find, like your Desktop. Or, if you have Git:

git clone https://github.com/ChimaOzonwoye/Via.git

Step 4: Set up the local program

Open a terminal and run the block for your system. If you unzipped somewhere other than the Desktop, change the first line to match.

Windows (Command Prompt):

cd Desktop\Via\server
python -m venv venv
venv\Scripts\activate
pip install -r requirements.txt

Mac / Linux:

cd ~/Desktop/Via/server
python3 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Step 5: Load the extension in Chrome

1. In Chrome, type chrome://extensions/ in the address bar
2. Turn on Developer mode (top right corner)
3. Click Load unpacked
4. Pick the Via/extension folder
5. Click the puzzle piece in Chrome's toolbar and pin Via

How to scan a block

Each time you use Via, first make sure Ollama is running (it starts itself on Windows and Mac; on Linux run ollama serve), then start the local program in a terminal:

Windows: cd Desktop\Via\server, then venv\Scripts\activate, then python server.py

Mac / Linux: cd ~/Desktop/Via/server, then source venv/bin/activate, then python server.py

Leave that terminal open. Then:

1. Mark the block. In Chrome, open Google Maps and drop into Street View at one end of the block you care about. Click the Via icon and press Capture next to point A. Walk the view to the other end of the block and capture point B. Press Scan this block.
2. Review the sample points. A new tab shows the points Via placed along your block and the years of imagery available at each. Untick any point you want to skip, choose which sides of the street to assess, and if you care about one specific building, press Focus on one building and drag a box around it. If Via warns that two views may not match, adjust them by hand or skip that point.
3. Run and read. Press Compare years. Via works through the points one at a time, showing progress like "view 4 of 12", and you can cancel anytime. Results appear as they finish: the block summary on top, then one card per point with the old and new views side by side. Press Save report to keep the findings as a file named like via-report-2026-07-13.json.

Honest limitations

• Some blocks and some years simply have no imagery, and Via will say so rather than fill the gap.
• Trees, parked trucks, and scaffolding can block a frontage, and Via marks those views "cannot assess" instead of guessing.
• Imagery from the late 2000s is blurry, so comparisons against it carry less confidence.
• Via normally follows the shape of the street when placing its points. If it cannot look that up it measures a straight line instead, says so on the review screen, and on a block that bends some points may then sit off the street.
• The vision model is occasionally wrong, which is why every card shows you both images so you can check its work.
• The imagery comes from an unofficial way of reading public Street View, so the tool may stop working if Google changes things; details in docs/limitations.md.

Documentation

• Limitations, what it cannot do, in plain words
• Prompt book, the exact question the vision model is asked and why
• Lineage, the research Via extends