| Name | Modified | Size | Downloads / Week |
|---|---|---|---|
| Parent folder | |||
| NetSentinel.exe | 2026-11-11 | 81.8 MB | |
| NetSentinel.exe.sha256 | 2026-11-11 | 83 Bytes | |
| NetSentinel.exe.sig | 2026-11-11 | 210 Bytes | |
| NetSentinel v1.15.0 source code.tar.gz | 2026-11-11 | 1.8 MB | |
| NetSentinel v1.15.0 source code.zip | 2026-11-11 | 1.9 MB | |
| README.md | 2026-11-11 | 4.4 kB | |
| Totals: 6 Items | 85.5 MB | 0 | |
2515 unit tests plus 14 integration checks. From 1.10.x or later this
installs from the Update button. A week of 1.14 on the owner's network raised
277 alerts for 62 distinct things, nearly all of them routine. This release
reports each thing once, says why it alerted, and adds a test
(test_detection_corpus.py) that replays the attacks each rule exists for: a
change that quiets a false alarm and an attack with it fails there.
In the first days after updating, each exposure already on your network is alerted once more as it is listed, and each routine upload is reported once while its history builds up.
- Exposures, kept apart from alerts. A plaintext login, an unencrypted service, a device answering SNMP's default community, software with a known exploited flaw or a bad certificate is a standing condition, not an event. 248 of the week's alerts were 43 of these repeated: Home Assistant's token over plain HTTP on four days, an energy monitor's Modbus port as CRITICAL.
- Each is now one entry per device and issue, in a new Exposures tab and dashboard page, with when it was first and last seen and how often.
- A new one raises one alert; a repeat only updates its entry. It alerts again only if it gets worse, or comes back after you mark it fixed. Accept one you have decided to live with.
- Exposures are rated at most HIGH. CRITICAL is for signs of an attack, and verification no longer raises a confirmed MEDIUM exposure to HIGH.
- Large uploads judged by their own history. The rule counted bytes per address and started again at 100 MB every day, so Tailscale to one peer, Chrome Remote Desktop and Claude were 13 HIGH alerts. It now keeps 30 days of uploads per program and destination (a service's many addresses share one history through the domain this PC's own DNS lookups gave them). A program's first large upload to a destination is reported; after that only a day past four times its biggest, and within a day at four times each level, not two.
- The check looks at the uploading program's signature. A signed program counts in its favour, an unsigned one against, a broken signature heavily against; general-purpose tools attackers upload with (curl, PowerShell, certutil, rclone and the like) get no credit for being signed.
- Volume counts only past 1 GB and 5 GB a day. A VPN to its peer is no longer a "LIKELY THREAT" because it moved 400 MB.
- Marking an upload Real holds that pair: it is reported from 100 MB again and learns nothing.
- Every alert says what it matched. "Suspicious User Agent found in unencrypted traffic" now names the user agent, what in it matched, and the site (with any token in the request hidden). One site behind several addresses is one alert. Card numbers, SSNs, cookies and parking credentials never reach an alert, only what is safe to show.
- Those five HIGH user-agent alerts were Edge running headless for its own background job, fetching from edge.microsoft.com. A headless browser asking its own maker is no longer flagged, and does not use up the once-per-source note, so automation to any other site still is.
- A verdict acknowledges, and each rule keeps a record. Real and False positive acknowledge the alert as well, in the dashboard and now in the desktop app's alert window, which had no way to judge an alert before. Rule record (desktop) and How each rule has done (dashboard) show per rule how many alerts were judged real or false: where detection needs work.
- A device answering SNMP's default community was missed when it was the first device the sender asked: its alert matched the sender's in the app's once-a-day check and was dropped as a repeat.
- New dashboard endpoints:
GET /api/exposures,GET /api/rules, and the token-gatedPOST /api/exposures/state?id=&state=open|accepted|fixed.
Installing. Install Npcap first (tick "WinPcap API-compatible
mode"), then run NetSentinel.exe. It asks for administrator rights because packet
capture needs them; from 1.14.2 the prompt names the publisher, James Cupps, since
the exe is code-signed. NetSentinel.exe.sha256 holds the checksum to verify the download,
and NetSentinel.exe.sig the signature the built-in updater checks before installing.