NORSESTAR
DarkWatch Platform · Valkyrie AI

Read the firehose.
Automatically.

Valkyrie is the AI layer of the DarkWatch platform. It reads 1.3 billion events a day so your analysts don't have to — clustering campaigns, flagging machine-generated attacks, and drafting the investigation before anyone opens a ticket.

94%machine-vs-human precision
6×faster first triage
2,400+campaigns clustered / day
<30 sto a drafted investigation
Signal, not noise

An analyst copilot for the AI era

Attackers now run on language models and autonomous agents. Valkyrie is built to recognize that — separating hand-run intrusions from machine-generated ones and handing your team a head start instead of a backlog.

  • Detects LLM- and agent-driven attack patterns
  • Auto-clusters related activity into named campaigns
  • Natural-language query across all DarkWatch telemetry
  • Drafts investigations with evidence and next steps
Request a demo ▸
Capabilities

What Valkyrie does for analysts

Every capability points at one goal: fewer windows open, faster decisions, and a clear line between human and machine adversaries.

Machine-vs-human scoring

Every cluster gets a calibrated verdict — hand-run, LLM-orchestrated, or fully autonomous agent — with the evidence behind it.

Named-campaign clustering

Related activity across sensors and time collapses into a single named campaign you can track as one object.

Natural-language query

Ask in plain English across all DarkWatch telemetry — no query language, no schema spelunking.

Drafted investigations

Valkyrie writes the first draft: timeline, affected assets, indicators, and recommended next steps for you to edit.

Inline enrichment

Pulls RuneFeed context and Yggdrasil pivots straight into the triage view.

Analyst edit history

Every Valkyrie suggestion is tracked and reversible, so the human stays accountable for the call.

How it works

From firehose to first draft

Four stages turn raw sensor telemetry into an investigation an analyst can act on.

1
Ingest

Read the firehose

Valkyrie consumes the full DarkWatch stream from the Bifrost grid — 1.3B events a day, enriched in flight.

2
Cluster

Group the activity

Related events across sensors and time collapse into coherent, named campaigns instead of isolated alerts.

3
Classify

Human or machine?

Each campaign is scored hand-run, LLM-orchestrated, or autonomous agent — with calibrated confidence.

4
Draft

Write it up

Valkyrie drafts the investigation — timeline, assets, indicators, next steps — for an analyst to review and own.

Valkyrie's detection and drafting models are served from NorseStar's internal model registry at models.internal.norsestarsecurity.com, maintained by the Threat Research team. Model cards and evaluation sets are available to Enterprise customers under NDA.

Editions

From one analyst to the whole org

Start with a single seat; scale to shared workspaces, custom classifiers, and private model access.

Analyst

Single seat

CustomPer analyst / year
  • Natural-language query over your DarkWatch data
  • Machine-vs-human scoring
  • Auto-clustered campaigns
  • Drafted investigations
Request a demo ▸

Enterprise

Org-wide

CustomAnnual subscription
  • Everything in Team
  • Private model registry access
  • Custom classifiers & tuning
  • SSO / SAML + audit logging
  • Dedicated research liaison
Request a demo ▸
Release notes

Valkyrie changelog

Currently shipping Valkyrie 2.3. Highlights from recent releases.

2.3September 2026

Classifier v4 & agent detection

  • Machine-vs-human classifier v4 adds agent-loop and tool-call pattern detection
  • Confidence calibration now surfaced per verdict
  • Drafted investigations cite Yggdrasil pivots inline
2.2May 2026

Natural-language query

  • Natural-language query across all DarkWatch telemetry
  • Auto-drafted investigation summaries with analyst edit history
  • RuneFeed enrichment inline in the triage view
2.1January 2026

Campaign clustering

  • Named-campaign auto-clustering across sensors and time
  • First machine-generated-attack detector (v1)
  • Private model registry support for Enterprise
The DarkWatch suite

Valkyrie works with the whole platform

Triage is one layer. These products feed Valkyrie and act on what it finds.

DarkWatch Live

The global attack map

Renders every attack the sensor grid sees, in real time, with AI-attack fingerprinting — the screen on the SOC wall.

Explore DarkWatch Live ▸

Bifrost Sensor Network

Global collection grid

Passive sensors, honeypots, and dark-IP listeners that capture the raw attack traffic Valkyrie reads.

Explore Bifrost ▸

RuneFeed Intelligence

Curated threat feeds & API

Enriched indicators and AI-attack signatures delivered as STIX/TAXII, a REST API, or native SIEM connectors.

Explore RuneFeed ▸

Yggdrasil Graph

Attribution & infrastructure graph

Connects every indicator — domains, certs, ASNs, wallets, operators — into the whole adversary tree.

Explore Yggdrasil ▸
Put the firehose on autopilot

See Valkyrie triage your traffic

Thirty minutes with a NorseStar intelligence engineer and a live look at how Valkyrie clusters and classifies what our sensors already see hitting your sector.

Request a demo ▸