Yes. By default your code never leaves your computer.
Concept Scanner reads the folder you choose and sends it to a model running on the same machine, through Ollama. Nothing goes to us. Nothing goes to a cloud model unless you choose one. And one flag makes that a hard rule.
Make it a guarantee.
Add --local-only to a scan and it refuses to run unless the model server is on this machine. It checks before anything is sent, and it also refuses an Ollama model that actually runs somewhere else.
From the terminal
concept-scanner scan ./client-project --local-only
What we saw when we tried it
We pointed a --local-only scan at a model server on another host. It refused before sending anything, named the host the code would have gone to, and left dropping the flag as a decision for us to type ourselves.
The flag is never saved, so give it on every run. In the window, Settings shows which model server a scan will use; the default is Ollama on this computer.
Where your code goes, by setup.
There are four ways to run a concept scan. The model is what decides where your code goes.
| Setup | Where your code goes | Good for |
|---|---|---|
| The app or terminal, with Ollama on this computer (default) | Nowhere. It is read and analyzed on this machine, and the results stay here. | Client code, unfiled work, anything confidential. |
| The app or terminal, with a cloud model you choose | To that provider, on your own key and under its terms. The scan shows its estimated cost before it spends. | A computer without the memory for local models, or a faster run. |
| The GitHub Action | To the provider in your workflow file, or nowhere beyond your own runner if it runs Ollama. | Scanning on every release, inside your own CI. |
| The agent skills | To your agent's model provider, like anything else the agent reads. | A quick look inside an agent you already use. |
An Ollama server on another machine counts as remote: your code goes to that machine. --local-only refuses it.
The other connections, all of them.
Update check, if you say yes
It asks once whether it may check for a newer release. With a yes, it asks GitHub at most once a day when a scan starts. It sends nothing about you or your code.
Model downloads
Ollama downloads the models the first time you use them. After that a local scan needs no network at all.
Scanning a GitHub URL
Give it a repository URL instead of a folder and it clones from GitHub first. A local folder needs no clone.
Usage reports: none
v4.1.0 sends no usage reports, to us or anyone. If a later release adds them, it will say so on first run, and this page will say so too.
Where your results live.
Scans, settings and any keys you save are in ~/.concept-scanner on your computer. Uninstalling the app keeps that folder, so your scans survive a reinstall; delete it yourself when a matter closes. The window is a page on your own computer: it listens only on 127.0.0.1 and refuses requests from any other machine or website.
Questions people ask.
Does anything go to ObviouslyNot?
No. There is no account, no sign-in and no server of ours in the path. The release you download from GitHub is the whole program.
Is a local scan enough for confidentiality?
It keeps the code off third-party servers, which is the part software can settle. Whether a workflow meets your duties to a client is your call; our page on cloud AI and patent privacy covers the frameworks, including the February 2026 ruling on AI chats and privilege.
What computer do I need for local models?
Memory is what matters: about 17 GB for the smallest profile, Fast, and 23 GB for Balanced. The download page has the full table, and the app recommends the profile your computer can hold.
What should I review once it runs?
How to review the results, on a real concept.