Veridex · Investigate
The investigation is done when the incident arrives
For each anomaly from Huntra, Veridex forms hypotheses, pivots across your data sources, pulls in threat intelligence and writes one explained incident with the evidence attached.
INC-0412 · Veridex
- Pivots run
- 37
- Sources queried
- 5
- Related anomalies
- 41
- Intel matches
- 2
- Time to narrative
- 3 m 40 s
Capabilities
What Veridex does
Plan
Hypothesis-driven
The agent decides what to check next, the way an experienced hunter would.
Pivot
Across every source
Moves between network, identity, cloud and endpoint data without an analyst copying IDs.
Correlate
Temporal attack graphs
Links entities and events over hours or weeks to catch low-and-slow campaigns.
Ground
Threat intel retrieval
Checks indicators and techniques against threat intel and your own case history.
Explain
Evidence-linked narrative
Every claim in the write-up links to the events that support it.
Defer
Knows when to ask
Low-confidence conclusions are flagged for a person, not presented as fact.
How it works
Inside Veridex
Receive
Take a scored anomaly and its entities from Huntra.
Investigate
Plan and run pivots across sources.
Correlate
Merge related activity into one incident graph.
Report
Write the narrative and hand it to Quellix.
Output
A write-up your incident lead can read in two minutes
Veridex writes in plain language, in order, with each step tied to its evidence and technique.
Example narrative · INC-0412
What happened
- 1At 02:14 the service account svc_backup signed in from 10.40.7.0/24, a subnet it had not used in 90 days of baseline.
- 2Three minutes later it opened the admin share on hr-fs-02 from fin-db-07 (T1021.002).
- 3At 02:21 a process on hr-fs-02 read credential-store memory (T1003).
- 4No rule fired. Huntra scored each step as unusual; Veridex linked them by identity and time.
Illustrative. In the product every sentence links to the events behind it.
Specification
Inputs, outputs and hardware
| Item | Detail |
|---|---|
| Inputs | Huntra and Sentix anomalies, raw telemetry for pivots, threat-intel feeds, case history |
| Output | Incident with narrative, evidence links, ATT&CK techniques, severity and confidence |
| Agent | NeMo-based planner with guardrails; retrieval with NeMo Retriever and cuVS |
| Graph | RAPIDS cuGraph temporal correlation |
| Hardware | H100 / H200 for model hosting and tuning, L40S / A100 for production inference |
Part of the loop
Where Veridex fits
Reads full telemetry at line rate and scores behaviour against per-tenant baselines.
02 · InvestigateVeridexPivots across sources and turns related anomalies into one explained incident.
03 · ContainQuellixChecks blast radius, then contains inside the guardrails you set.
04 · EdgeSentixBrings the same hunting to OT, IoT and air-gapped networks.
05 · SimulateSynthraEmulates attacks to train the models and test coverage.
Questions
About Veridex
Can it make things up?
Veridex only states what it can link to evidence. Unsupported steps are marked as hypotheses, and low-confidence incidents go to an analyst for review.
Does it replace our analysts?
No. It does the pivoting and write-up so analysts spend their time deciding what to do.
Can we see how it reached a conclusion?
Yes. Each incident includes the agent's investigation trail: the questions it asked and the data it looked at.
Point Veridex at your own telemetry
We'll scope an assessment with you: which sources {name} would read, for how long, and what we'd report back.