Windows security posture · open source
Find out how exposed
your Windows fleet
actually is.
EndpointRadar audits antivirus, firewall, patching, open ports, running services and installed software — then scores the risk, explains it in plain English, and maps every finding to NIST CSF 2.0. No agent to deploy. No server to run. No licence to buy.
The problem
Commercial EDR tells you when you're attacked.
Nothing tells you how easy you'd be to attack.
Most small teams, labs and classrooms have no visibility into basic endpoint hygiene. Is Defender actually on? Is the firewall off on the public profile? Is SMB listening? When was this box last patched? EndpointRadar answers those questions in one command and gives you a number you can track over time.
Antivirus
Detects installed AV products through WMI SecurityCenter2, plus real-time protection state — not just "is Defender installed".
Firewall
Checks Domain, Private and Public profiles independently, because one profile being off is the whole gap.
Patching
Reads the hotfix history, works out days since the last update, and flags suspiciously thin patch counts.
Exposure
Enumerates listening ports and running services, and singles out the ones that widen the attack surface — 445, 3389, RPC.
Four ways to look
One engine, four vantage points
Every mode produces the same record shape and runs through the same scoring code, so a local scan, a remote scan and a network sweep are directly comparable.
The deep, authenticated view
Runs every check against the machine you're sitting at. This is the richest view: AV state, firewall profiles, hotfix history, listening ports, running services and the full installed-software inventory from the registry.
- NeedsNothing but Python. Run as admin for full service and port visibility.
- ProducesA complete scan record plus a risk score from 0 upward.
- Good forAuditing your own box, CI gates, teaching, baselining.
endpointradar scan endpointradar scan -f json -o box.json endpointradar scan -f text # long report endpointradar scan --fail-on high # exit 2 in CI
Agentless, over WinRM
Stages the stdlib-only subset of the scanner onto the target over WinRM, runs it with the target's own Python, reads the JSON back and deletes nothing you'd miss. The remote host needs no pip packages and no permanent install.
- NeedsWinRM enabled, admin credentials, Python on the target.
- CredentialsRead from an env var or prompted for — never passed as an argument.
- Good forAuditing a handful of machines you administer, without deploying anything.
$env:ENDPOINTRADAR_PASSWORD = '...' endpointradar remote 10.0.0.51 -u Administrator # or just let it prompt you endpointradar remote 10.0.0.51 -u .\Admin
Who is even on this network?
Sweeps a subnet in parallel, resolves hostnames and NetBIOS names, reads the ARP table for MAC addresses, and classifies each device as a Windows PC, a mobile device or unknown using OS hints, hostname patterns and MAC vendor prefixes.
- NeedsNothing. No credentials at all.
- ProducesAn inventory of responding hosts with type and vendor attribution.
- Good forFinding the devices you forgot were there.
endpointradar discover endpointradar discover 192.168.1.0/24 endpointradar discover --workers 64 -f json
The attacker's view
Port-scans every responding device for SSH, HTTP, HTTPS, SMB and RDP and scores what an unauthenticated attacker on the same network could see. Results are explicitly labelled limited visibility so they're never confused with a full authenticated scan.
- NeedsNothing. No credentials at all.
- WeightingSMB 8 · RDP 5 · SSH 3 · HTTP/S 2 — exposure, not vulnerability.
- Good forSeeing your network the way someone who just joined the Wi-Fi does.
endpointradar sweep endpointradar sweep 192.168.1.0/24 endpointradar sweep -f json -o exposure.json
Real output
What a scan actually gives you
This is rendered live in your browser from
examples/sample-scan.json — the same anonymised record
shipped in the repository and used by the test suite. Nothing here is mocked up.
—
—Risk drivers
Findings
Command reference
The whole CLI
Nine subcommands. Every one takes -f json for machine-readable output and
-o FILE to write it somewhere. -q and --no-color
work on either side of the subcommand.
| Command | What it does | Key options |
|---|---|---|
scan | Full posture scan of this machine | --fail-on --no-save -f summary|json|text |
discover | Find live devices on a subnet, no credentials | --workers --timeout --include-offline |
sweep | Unauthenticated port and exposure scan | --no-save -f table|json |
remote | Agentless scan of a remote host over WinRM | -u --password-env --timeout |
threats | NIST CSF 2.0 threat analysis of a scan | --max -v -d DEVICE |
report | Word report — fleet summary or per-device threats | fleet|threat -o out.docx |
show | Re-read a saved scan without rescanning | --all -v --fail-on |
dashboard | Launch the Tkinter desktop dashboard | needs the gui extra |
serve | Read-only web dashboard and JSON API | --host --port |
Use it as a CI gate
--fail-on makes the process exit with code 2 once the risk level reaches
your threshold, so a Windows runner can fail the build when a machine drifts out of policy.
- name: Endpoint posture gate
run: |
pip install endpointradar
endpointradar scan --fail-on high --no-color
shell: pwsh
Under the hood
How a finding becomes a number
Scoring is deliberately transparent and deterministic. There is no model, no cloud call, and no
hidden weighting — you can read the whole thing in risk.py and threats.py.
Collect
Native Windows tooling only — WMI, PowerShell, netsh, the registry, psutil. Every collector degrades gracefully when a source is unavailable.
Normalise
Everything lands in one JSON record shape, whether it came from this machine, a remote host or a credential-free port scan.
Score
Six components — antivirus, firewall, patching, ports, services, third-party software — each capped at 100, then weighted into contributions.
Explain
Findings become threats with likelihood × impact (1–5 each), a NIST CSF 2.0 function, and a fix written for a human.
Getting started
Installed and scanning in under a minute
The core install is one dependency. Everything heavier — WinRM, Word reports, the GUI, the web server — is an optional extra you add only if you want it.
Install
git clone https://github.com/4ryanwalia/EndpointRadar cd EndpointRadar py -m venv .venv .venv\Scripts\activate pip install -e . endpointradar scan
Optional extras
pip install -e ".[remote]" # WinRM scanning pip install -e ".[report]" # Word + Excel pip install -e ".[gui]" # desktop dashboard pip install -e ".[web]" # web API pip install -e ".[all]" # everything