Documentation Index
Fetch the complete documentation index at: https://quintsecurity.mintlify.app/llms.txt
Use this file to discover all available pages before exploring further.
Intelligence
Quint’s intelligence layer is the reason the platform sees more than a passive EDR. Two systems work together:Behavioral Intelligence (Tier 1)
Per-agent statistical scoring. Every tool call is scored against the agent’s behavioral baseline, flagged deviations trigger escalations, and threat signatures are distilled into fleet-wide rules. Pipeline:- Scoring pipeline — 4 gates (deny list, capability policy, rate limit, behavioral Bloom) + Gate 5 (corroboration)
- Agent fingerprint — stable identity across sessions, derived from model + platform + tool inventory
- Session relationships — parent/child spawns, model divergence, sub-agent detection
- Baseline floors — minimum signal required before an agent can be scored
- Threat signatures — distilled patterns pushed to every proxy in the fleet
Graph Intelligence (Tier 2)
Structural pattern detection that Tier 1 can’t express. Actions are materialized as a property graph in Memgraph, and a Variational Graph Autoencoder (VGAE) scores session subgraphs for anomaly. Pipeline:- Action graph — nodes: Action, Session, Agent, Resource; edges: NEXT, BELONGS_TO, STARTED_BY
- GNN features — 143-dim feature vector per action (capability, temporal, rule firings)
- VGAE anomaly score — reconstruction error normalized to [0, 1]
- Signature distillation — high-anomaly sessions → structural
ThreatSignature→ fleet push
How they compose
Tier 1 handles the common case fast (single-event decisions in under 35ms). Tier 2 runs asynchronously on session graphs for anomaly detection that catches novel patterns. When Tier 2 finds something interesting, it distills the pattern and pushes it back to Tier 1’s rule matcher — closing the feedback loop.Where to go next
Behavioral Pipeline
The 4-gate + Gate 5 scoring flow
Graph Schema
Nodes, edges, and ingestion
Threat Signatures
How patterns distill and propagate
Scoring Algorithm
VGAE + rule engine composite score