A security analyst on every server
It watches your macOS, Windows and Linux machines around the clock and answers your questions with the evidence. When a Windows server is under attack, block the attacker from your phone until you get there.
Free for 2 machines100 welcome creditsNo card
- list_processesunsigned_only=true
- outbound_connections
- list_autorunssuspicious_only=true
- recent_changeshours=24 executable_only=true
- event_log
One process stands out. nodeh (pid 4312) is unsigned, runs from /tmp/.cache/nodeh and holds a connection to 203.0.113.5:443. A launch agent written at 01:23 starts it at every login.
Not checked: the event log, because the agent on this machine is not elevated.
Illustrative data: fictional machines, people and addresses.
checks, compiled into the agent
emergency actions, each one approved by a person
inbound ports opened on your machines
to install, enrol and connect the agent
Every answer shows its evidence
The checks run on the machine, and the answer quotes what came back and says what it could not see.
Checks run in parallel
One question can start several checks at once: processes, sockets, autoruns, logons and recent changes.
Gaps are reported
If a check needs root, a list is cut short or the machine goes offline, the answer says so.
You set the severity
Save an answer as a finding, set its severity, assign it, and close it when it’s handled.
| Process | Remote | Signed |
|---|---|---|
| nodeh 4312 | 203.0.113.5:443 | unsigned |
| Safari 811 | 198.51.100.24:443 | system |
| nsurlsessiond 402 | 198.51.100.77:443 | system |
| Mail 790 | 192.0.2.41:993 | system |
Unsigned binary in /tmp holding an outbound connection
pid 4312 runs /tmp/.cache/nodeh and holds 203.0.113.5:443. The launch agent that starts it was written 40 minutes before the process started.
Illustrative data: fictional machines, people and addresses.
Running in about a minute
Install one binary, enrol it with a token from your console, and ask.
Install and enrol
One download for macOS, Windows or Linux; the installer checks its published SHA‑256. Enrolling needs a token from your console, so a leaked install command can’t add machines.
Connect it
It connects out to guardyn.dev and stays connected. Nothing on the machine listens for connections. Run as a service on a paid plan, it watches the machine as things happen.
Ask
Pick a machine and a model, and ask in plain language.
Act before you can reach the server
With guardyn actions, an owner or admin can act from Slack, the Investigate chat or the console: block an attacker, stop a hacked site or sign an intruder out. The agent on that server does it within seconds, and undoes a block by itself when its time is up.
42 failed sign-in attempts over Remote Desktop in 5 minutes, from 1 address
Someone is guessing at sign-ins from 203.0.113.45. Every attempt so far failed.
203.0.113.45 is blocked on WEB-01 until 15:03 UTC
Windows Firewall now blocks it in and out, and 3 open connections from it were ended. The agent removes the block by itself at 15:03 UTC, even if it cannot reach guardyn.
Illustrative data: fictional machines, people and addresses.
Off until you turn it on
Each server’s administrator turns emergency actions on at that server with one command, and can allow only some. guardyn cannot turn them on, and nothing it sends can.
A person approves each one
Only an owner or admin of your workspace can take one, after a confirm that says exactly what will happen. Nothing is taken automatically.
Temporary, and on the record
A block, a stopped site or a disabled account lasts the time you choose, at most a day at a time, and the agent undoes it even if it cannot reach guardyn. Each one is logged on the server and in your workspace.
Block an address
Blocks one IP address in and out on the server's own firewall and ends its open IPv4 connections. Undone by the agent when the time is up.
Stop a site
Stops one website: the IIS site, with the application pools only it uses. Started again by the agent when the time is up.
Move a file into quarantine
Moves one file out of a web folder into the agent's quarantine on the server, with its fingerprint kept. Nothing is deleted.
Sign an account out
Ends every Remote Desktop and console session the account has open on the server. It can sign in again; disable it to stop that.
Disable an account
Stops one of the machine's own accounts signing in at all, by password or by key. Enabled again by the agent when the time is up.
Stop a process
Stops one running program, after the agent checks it is still the same program. There is nothing to undo.
Everything it can do is compiled in
Its checks and its six emergency actions are built into the agent. Our server can ask for one of them by name, and for nothing else.
What the agent can never do
- Run a commandNothing it is sent becomes a command, a script or a program, and there is no shell to switch on.
- Turn on emergency actionsOnly an administrator at the machine can, with one command there. Nothing guardyn sends can.
- Listen on a portIt only connects out, and checks and actions come down that one connection.
- Read any file it likesFile contents come back only from a short list of forensic paths, and keys,
.envfiles and credential stores are always refused.
It keeps its own record
Every check and every emergency action is written to an audit log on the machine, so the machine has its own account of what was looked at and what was changed.
If our server were breached
An attacker could see whatever the agent’s checks return from your machines. On one with emergency actions turned on, they could also take those, within the agent’s own guard rails. They could not run a command of their own, or turn actions on anywhere.
What it can check
32 checks in 8 groups; each machine offers the ones its operating system supports.
What is running
Processes, their ancestry, their signatures and the files they hold open; services, kernel modules and drivers.
list_processes process_tree process_open_files list_services loaded_modules
What is reachable
Listening ports, live connections, where they go, and what resolves them.
list_sockets outbound_connections resolve_host
What is using the machine
CPU and memory by process and by program, the database queries behind a spike, who sent them, and why one app is slow.
resource_usage database_activity app_activity
What survives a reboot
The persistence surface: launchd, Run keys, services, cron, WMI and shell profiles, plus recently installed software and containers.
list_autoruns installed_packages list_containers
Who can get in
Local accounts, administrator membership, recent logons and failed attempts, and certificates someone added to the trust store.
list_users recent_logons list_certificates
What changed
Recent filesystem activity, executables in writable places and file hashes, and the files themselves: find one, inspect it, list a folder, read a config file.
recent_changes find_files hash_files inspect_path list_dir read_text
What the defences say
Antivirus, firewall, disk encryption, SIP, Gatekeeper and Secure Boot; integrity checks, a CIS-aligned benchmark, the event log, and one scan that runs them all.
security_posture cis_benchmark check_integrity event_log full_scan
What the browser has been doing
History, downloads, cookie hosts, and installed extensions with the permissions they hold. Metadata only, never a cookie value.
browsing_history browser_downloads browser_cookies list_browser_extensions
checks in all, up to 31 on any one machine.
Choose a model, set spending limits
Pick the model for each conversation, and cap what each workspace and person can spend.
Pick the model
Higher plans unlock more capable models: fast on Free, balanced on Starter and frontier on Team.
- Claude Sonnet 5Recommendedbalanced
- Claude Opus 5Deepestfrontier
- Claude Haiku 4.5fast
OpenAI, Google and OpenAI-compatible models will be listed once we turn them on.
Cap the spend
Each question is checked against your balance before it starts, and each model call is charged at its exact cost. One credit is one US cent at retail model prices: the provider’s list price × 3.
A typical question on the balanced tier uses 8 to 25 credits.
Free for two machines, paid plans for more
Every plan includes every check, and every paid plan adds the 24/7 watch. Plans differ in credits, machines, seats and model tiers, and Team and Enterprise add guardyn actions.
Showing monthly prices
Free
Try it on your own machines
+100 welcome credits
Start free- 200 credits a month
- 2 machines, 2 seats
- Fast models
- 1 investigation at a time
Starter
For a small fleet
or $470 a year
Choose Starter- 5,000 credits a month
- 15 machines, 5 seats
- Balanced models
- 2 investigations at once
- 2 API keys
- Real-time 24/7 watch
TeamPopular
For a security team
or $1,910 a year
Choose Team- 25,000 credits a month
- 100 machines, 20 seats
- Frontier models
- 5 investigations at once
- 10 API keys
- Audit-log export
- Real-time 24/7 watch
- guardyn actions on Windows
Enterprise
For large fleets
Terms agreed with you
Contact sales- 150,000 credits a month
- 5,000 machines, 500 seats
- 20 investigations at once
- 50 API keys
- Audit-log export
- Real-time 24/7 watch
- guardyn actions on Windows
Questions
Can the agent be used to attack my machines?
It cannot run a command it is sent: no check or emergency action takes one. Its checks look at processes, ports, files, accounts and logs, and none of them changes anything. The six emergency actions stay off until that machine's administrator turns them on there, and an owner or admin approves each one. If guardyn itself were compromised, it could run those checks, and take emergency actions only where they are already on, within the agent's own guard rails.
Can it stop an attack before I get to the server?
For a while. On a plan with guardyn actions, an owner or admin can act from Slack, the Investigate chat or the console: block the attacker's address, stop the hacked site, sign the intruder out, disable an account or stop a process. This works on a Windows server whose administrator turned emergency actions on. A block, a stopped site or a disabled account lasts the time you chose, and the agent undoes it by itself.
Do you train on my data?
No. Evidence from your machines goes to the model provider you selected, to answer your question, and nowhere else. Nothing is used to train anything.
What happens when I run out of credits?
Investigations stop with a message saying how many credits the next one needs. Nothing is deleted, and your agents stay connected.
Is there a minimum term?
No. You pay for a month or a year at a time through PayPal. Nothing renews automatically; three days after paid time ends, the workspace moves to Free and nothing is deleted.
Where does it run?
guardyn.dev is a hosted service. You install the agent; everything else runs on our infrastructure, so there is nothing for you to operate, patch or back up.
Which operating systems are supported?
macOS, Windows and Linux, with one agent binary per platform. So far, only the Windows agent has emergency actions.
Your first machine takes a minute
Free for 2 machines. No card required.