Not just what threats exist — but who they matter to

Black Swarm
Autonomous Agentic Threat Intelligence System

Three-Axis Intelligence / Evidence-Grounded / Always On
01

Who, Not What

Every CTI tool tells you what threats exist. Black Swarm tells you who they matter to. When a zero-day drops, the system instantly maps relevance across all monitored organizations — by industry, tech stack, geography, and exposure surface. The question shifts from “is this threat important?” to “who should I care about right now?”

02

Customer Digital Twin

For each organization monitored, Black Swarm maintains a continuously evolving digital twin — a persistent model of their technology stack, business exposure, regulatory posture, and threat surface. Auto-discovered from Shodan, DNS, certificates, and enriched over time. No competitor does this today.

03

Evidence, Not Assertion

Every AI-generated claim links to traceable evidence, standardized MITRE ATT&CK tags, and confidence scores. The system proposes; the analyst decides. Hallucination is treated as an engineering problem, not a marketing footnote — with RAG grounding targeting <2% unsubstantiated claims.

How It Works

Six Layers / Raw Signal to Intelligence

Black Swarm operates a continuous 6-layer pipeline. Sources are discovered, validated, and collected from across the open web, social platforms, and dark web. Raw evidence is sanitized, enriched by hybrid LLM inference, synthesized into analyst-ready intelligence, and presented through three distinct axes.

L0
Discover
4-tier source discovery engine with LLM scoring and human approval gates
L1
Collect
OSINT feeds, social platforms, dark web — 3 categories, scheduled polling
L2
Sanitize
NVIDIA NeMo Guardrails: multi-pass filtering, IOC extraction, trust boundary enforcement
L3
Enrich
Multiple LLMs — the right model for the right task, from Claude Opus 4.7 to Llama 3.1 70B
L4
Synthesize
Cross-source correlation, narrative generation, customer impact scoring, spike detection
L5
Present
Three intelligence axes: Global, Customer, and Investigation workspaces
Explore Intelligence Journey
User Journey

From Global Noise to Local Action

Black Swarm reshapes the analyst's workflow from reactive alert triage to proactive, evidence-backed investigation.

01

Intelligence Briefing

Open the dashboard to the Global Intelligence View. The overnight intelligence feed shows synthesized threat cards — each grounded in evidence from OSINT, social, and dark web sources. The weekly threat briefing summarizes the landscape. Threat Pulse KPIs surface severity distribution at a glance.

Global Feed / Threat Pulse / Weekly Briefing / Actor Activity
02

Threat Surfaces

A critical zero-day drops. Black Swarm auto-maps it against every customer digital twin in real time. The Customer Intelligence view lights up: “3 of your 47 customers are exposed.” Click to see which ones, why, and what their specific exposure vectors are — tech stack, geography, and attack surface overlap.

Digital Twin / Impact Matrix / Exposure Scoring / Auto-Mapping
03

Deep Investigation

Drill into the most-affected customer. Pivot through the Neo4j relationship graph — actors, campaigns, infrastructure linkage. Build a hypothesis in the investigation workspace. Query evidence via natural language. Generate an evidence-grounded report with MITRE ATT&CK mapping and export as STIX 2.1.

Graph Pivot / NL Query / Evidence Timeline / STIX Export
Architecture

Built for Autonomy / Governed by Trust

Black Swarm runs on purpose-built infrastructure with dedicated trust zones, persistent storage, and multi-model LLM inference. Every component is designed for persistence, auditability, and controlled autonomy.

LangGraph Orchestrator

Stateful agent orchestration with persistent checkpointing. Investigations survive restarts, shift changes, and infrastructure events.

Multi-Model LLM Intelligence

The right model for the right task — from Qwen2.5-14B for high-volume triage, through Llama 3.1 70B for synthesis, to Claude Opus 4.7 for deep reasoning. Continuous quality monitoring across all models.

Neo4j Knowledge Graph

Relationship graph of actors, campaigns, TTPs, and infrastructure. Cypher-powered pivot and hunt across entity relationships.

PGVector Semantic Search

High-dimensional embeddings for evidence similarity. Powers semantic clustering, deduplication, and RAG retrieval at synthesis time.

NVIDIA NeMo Guardrails

Multi-pass content sanitization on all raw evidence. Pattern-based and semantic filtering with escalated scrutiny for hostile-zone content.

Langfuse Observability

Full LLM trace logging, cost tracking, and quality metrics. Grafana dashboards and Prometheus alerting for platform health.

Graphiti Temporal Graphs

Temporal knowledge graph for tracking how threat actor relationships and campaign patterns evolve over time.

View System Architecture