Autonomous threat hunting

Hunt what your alerts miss

Lairhunt reads full network, identity, cloud and endpoint telemetry on GPUs, finds behaviour no rule fired on, and hands your analysts one explained incident instead of forty alerts.

Built on NVIDIA Morpheus, RAPIDS and NeMo. Runs in our cloud, your VPC or on-prem.

Example incident · INC-0412Threat · open

Lateral movement from fin-db-07

Techniques
T1078 · T1021.002 · T1003
Alerts folded in
41 → 1 incident
Entities
3 hosts · 2 identities
Severity / confidence
High / 0.92
  1. Login by svc_backup from a subnet it has never used
  2. SMB admin-share access, fin-db-07 to hr-fs-02
  3. Credential-store memory read on hr-fs-02
  4. Veridex links all three into one incident
Recommended: isolate hr-fs-02, disable svc_backupNeeds approval
Illustrative values. Real incidents link every step to the underlying telemetry.

What we are building toward

Fewer, better incidents

These are design targets for the platform, not measured customer results.

Telemetry inspected100%

Design target: full-fidelity flows, identity and log events, not a sampled subset.

Alerts to incidents41→ 1

From the example incident above. Related signals arrive as one case.

Time to explained incident< 5min

Design target from first anomaly to a written, evidence-linked narrative.

Containment actions100%

By design, every Quellix action is logged, explained and reversible.

The problem

Your SOC is triaging alerts its rules already chose

Most security stacks only look at events a rule matched. Analysing every flow, login and log line is too expensive on CPU pipelines, so slow and unfamiliar attacks sit in the noise for weeks.

Analysts spend their shift closing alerts by hand and re-deriving the same story each time. The intrusions that matter never raised an alert in the first place.

  • Rules only find what someone already wrote a rule for
  • Low-and-slow campaigns stay under every threshold
  • Investigation is manual, slow and inconsistent between analysts

Typical alert queue

  1. 01Impossible-travel login, closed as VPN
  2. 02Unusual SMB volume, closed as backup job
  3. 03Credential-store read, never alerted
  4. 04New admin share, closed as change ticket

Illustrative. Each line looked harmless alone. Together they were one intrusion.

How it works

From raw telemetry to contained threat

  1. Minutes to connect

    Stream everything

    Connect network taps, identity providers, cloud audit logs and EDR. Huntra reads them raw.

  2. Line rate

    Hunt continuously

    GPU pipelines score every event against per-tenant baselines and map anomalies to ATT&CK.

  3. Minutes

    Investigate automatically

    Veridex pivots across sources, correlates related activity and writes the incident.

  4. Your policy

    Contain with guardrails

    Quellix recommends or runs reversible actions within the autonomy level you set.

Built on the NVIDIA stack

GPU-native from packet to playbook

Full-telemetry hunting is only affordable on accelerated compute. Lairhunt is built for it end to end.

MorpheusCybersecurity data pipelines
RAPIDSGPU dataframes, ML and graphs
NeMoInvestigation agent and guardrails
Triton and NIMModel serving
TensorRTLow-latency scoring
BlueField DPUIn-path capture and enforcement
Jetson and IGXSentix edge appliances
NVIDIA AI EnterpriseSupported deployment

See every integration →

For engineers

Incidents arrive where your team already works

Every incident is available through the API and as a signed webhook. Route it to your SOAR, ticketing system or chat with the evidence attached.

  • REST API and signed webhooks
  • Splunk, Sentinel, Elastic and Chronicle connectors
  • CrowdStrike, Defender, Okta and Entra ID response actions
incident.opened · example payload
// POST to your SOAR or ticketing webhook when an incident opens
{
  "event": "incident.opened",
  "incident": "INC-0412",
  "title": "Lateral movement from fin-db-07",
  "severity": "high",
  "confidence": 0.92,
  "techniques": ["T1078", "T1021.002", "T1003"],
  "entities": { "hosts": 3, "identities": 2 },
  "recommended_actions": [
    { "action": "isolate_host", "target": "hr-fs-02", "reversible": true },
    { "action": "disable_identity", "target": "svc_backup", "reversible": true }
  ],
  "url": "https://console.lairhunt.com/incidents/INC-0412"
}

Questions

Frequently asked

Do we have to replace our SIEM or EDR?

No. Lairhunt reads the telemetry you already collect and sends incidents back to the tools you use. Many teams keep their SIEM for retention and compliance.

Will it take containment actions on its own?

Only if you allow it. Quellix starts in recommend-only mode. You choose which actions, if any, run without approval, and every action can be rolled back.

Where does our telemetry go?

It stays in your tenant, in the region you choose, or on your own hardware. Cross-tenant learning uses only derived, anonymised behaviour patterns you opt in to share.

Is Lairhunt available today?

We are working with a small group of design partners. Join the program to run Lairhunt on your own telemetry.

See what Lairhunt finds in your telemetry

Bring a week of flow and identity logs. We'll show you the incidents your current rules did not raise.