Whisper Graph XDR
Give every machine and agent an identity.
Contain it from the network, not the box.
Each machine, container, AI agent and device gets its own public IPv6 address on Whisper's network. Every lookup and connection it makes is checked against the internet graph and logged with who is on the other end. When something goes wrong, you cut it off from outside, even if the host is owned.
A host-only tool trusts the host to report on itself.
Once an attacker owns a machine, everything on it answers to them, including the agent meant to contain it. And the IP lists your policy relies on are out of date by the end of the week.
Four steps, and the first one needs no install.
Give it an identity
A /128 on AS219419, with signed reverse DNS, a pinned certificate and a public registry record. Anyone can check it with dig and curl.
Connect it
Route all its traffic through a tunnel, send chosen apps through our egress proxy, or use only our DNS.
Check every connection
Each lookup and connection is scored against the graph and checked against your rules as the name resolves and the packet leaves.
Contain from outside
Revoke the identity or flip its firewall to deny. It works even when the host no longer answers.
$ dig -x 2a04:2a01:c787:fa3b:d78c:23bd:a6:c4ae +short
ad78c23bd00a6c4ae.tb00ab91b62bf1dbae50af8a9a292633e.agents.whisper.online.
$ curl -s https://rdap.whisper.online/ip/2a04:2a01:c787:fa3b:d78c:23bd:a6:c4ae | jq .status
[
"active"
]Every connection, with who is on the other end.
A log line that says "443 to 104.21.x.x" tells you nothing. Graph XDR logs the destination with its owner, network, country and risk, the moment it appears.
Activity and egress
Every DNS answer and connection per identity, with the policy decision and a graph score.
The map
Where your fleet's traffic lands, placed by the graph, with risky destinations marked.
Your fleet's landlords
The organisations that own the networks your fleet reaches, with their abuse contact and how much listed activity their space carries.
Campaigns
Named malware families and their live infrastructure, so you can see whether your fleet touched any of it.
See who is looking for your machines.
Before an attack there is reconnaissance: someone resolves your hosts, looks up their registry records or walks the reverse zone. That traffic never reaches the host, so no host sensor can see it. Graph XDR does, because the lookups arrive at our DNS.
Rules written in what is behind the address.
Rules run at the resolver and at egress, against the live graph. Start in monitor mode, watch what would have been stopped, then enforce.
- Owners and networks
Anything run by an organisation or network you name, wherever its addresses move.
- Live
- Countries
Deny a list of countries, or allow only a list, counting every country a name lands in.
- Live
- Bulletproof hosting
Anything hosted on a known bulletproof network, even when the address itself is clean.
- Live
- Tor exits
Anything that resolves onto a Tor exit.
- Live
- Sanctions
Names, addresses and networks on a sanctions list.
- Live
- RPKI-invalid
Destinations reached over a route that fails RPKI validation.
- Live
- New registrations
Domains registered in the last few weeks.
- Live
- Firewall
A per-identity firewall, default allow or default deny, with rules by address, network, host and port.
- Live
Every change is staged, reviewed as a before and after, applied through an audited change, and recorded with who made it.
If our graph is unreachable, resolution fails open by default. A strict mode that blocks instead is available.
The host's side of the story, scored by the graph.
The network layer covers every identity. The sensor adds what runs on the machine: processes, files and connections joined into one record, with the graph’s verdict on every destination.
What it watches
Process starts, file changes and network activity, joined into one causal chain: write, then run, then connect.
What it detects
Ransomware behaviour, process injection, download-then-run chains and Sigma rules. Detections are held for your decision, never acted on silently.
What it can do
When you arm it, stop a process, deny an execution or quarantine a file on the host. Every result says it came from the sensor.
- Linux
Kernel level, with an enforcement path that can deny an execution. Installed in one line.
- Live
- Windows
User mode, no kernel driver, deployable through Intune.
- Live
- macOS
Connection telemetry today. The deeper Apple feed follows.
- In development
Debian 12, published test run of 27 August 2026. Telemetry is visibility, not a detection rate.
Containment the host cannot veto.
- Revoke
The identity stops resolving, its firewall drops traffic, and its public record is withdrawn.
Network, off-host
- Default deny
Every flow not explicitly allowed stops at our edge.
Network, off-host
- Stop a process
The sensor ends a running process on the machine.
With the sensor
- Quarantine a file
The sensor moves a file into quarantine and checks its hash.
With the sensor
See the blast radius first: revoking a node cuts everything connected beneath it.
Confirm, commit, then watch the public record change with a live dig. Every action goes to the audit trail, and every action can be reversed.
Evidence your auditor can re-run.
Controls with evidence
Mapped to SOC 2, ISO 27001, PCI DSS, HIPAA and CMMC. Each cites a live count or a check you can run now, and says where a control is still partial.
Exports with a digest
CSV and JSON exports carry a SHA-256 digest, so anyone can confirm they were not edited.
A tamper-evident log
Every identity granted or revoked is written to a public, append-only log, cosigned by a witness.
Exposure that counts what runs
Vulnerabilities ranked by whether the package runs and faces the network, not by raw CVE count. Linux hosts today.
A private mesh made of public addresses.
Connect your servers and agents to each other over their Whisper identities. One document decides who may reach whom, and a rule can check the graph, such as refusing a destination whose route fails RPKI.
One XDR for every machine you run.
Network and host in one record, for every kind of identity. What joins it, and what the sensor adds where it runs.
| How it joins | On the host |
|---|---|
| Servers | |
| One line on Linux. Intune or the universal installer on Windows. | Linux sensor, Windows sensor |
| Laptops and desktops | |
| A signed profile from Intune, Jamf, Kandji, Omnissa or Hexnode. | Windows sensor, macOS in development |
| Containers | |
| An identity per container. | Network layer |
| AI agents and sub-agents | |
| One command for supported frameworks, an SDK for serverless and edge, and an MCP server an agent can use to manage its own identity. | Network layer |
| Browsers | |
| Whisper Guard, from the Chrome Web Store and Firefox Add-ons. | Checks every site's name |
| OT and IoT devices | |
| The resolver-only tier. Nothing is installed on the device. | Network layer |
Your industry
Editions for connected vehicles, energy, telecom, OT and more.
Your clients
Partners run every client’s fleet from one portal, with branded reports.
What people ask first.
Give your first agent an identity.
Five identities and one seat are free, for good. After that, €3.99 per endpoint per month, billed yearly.