Threat intelligence engine

Meet RoCi.
She analyzes your
DNS patterns.

RoCi is Nantevo's AI-powered threat intelligence engine. She analyzes the DNS telemetry stream to detect behavioral anomalies, identify attack patterns, and automatically update threat rules — across every authenticated client, continuously.

RoCi never touches a live query. All analysis happens asynchronously on the log stream — after responses are delivered. DNS resolution latency is a hard constraint. AI inference is not. This design keeps both fast and accurate.

roci — async threat analysis pipeline
roci> ingest telemetry --stream
source ......... nantevo telemetry pipeline
clients ........ 247 active · analyzing
query responses already delivered to devices
 
roci> score --client eng-0x4f2a
baseline ........ 2,400 queries/day
window .......... last 60 minutes
✓ volume ........ within normal range
✓ domains ....... known good majority
 
! anomaly — 14:32:07 UTC
domain: xk92mf7a3n.cdn-analytics.net
entropy: 4.2 · dga pattern: probable
→ block rule pushed to threat feed
→ future queries to this domain: blocked
→ incident #4471 · client notified
 
roci>
Architecture principle

Zero latency impact.
The analysis happens after the answer.

DNS resolution speed is non-negotiable. RoCi's design separates the response path from the analysis path — queries are resolved and returned without waiting for any AI inference. RoCi works on what already happened, not what is happening right now.

Synchronous — query path

Response delivered instantly

Every step in the live query path is a fast lookup, never an inference.

  • Credential validation at proxy
  • Threat feed lookup — known-bad domains
  • Per-client policy enforcement
  • DNS resolution
  • Encrypted response returned
response: <50ms typical
response
delivered
Asynchronous — after response

Metadata logged to telemetry

Query metadata is written to the telemetry pipeline after the response is already on its way back to the device.

  • Timestamp and client identity
  • Response code and latency
  • Domain (if content logging enabled)
  • Threat classification from feed
logging: async · non-blocking
stream
ingested
RoCi — intelligence layer

Analysis on the log stream

RoCi consumes the telemetry stream, runs behavioral scoring, detects patterns, and pushes rule updates back into the synchronous threat feed.

  • Behavioral baseline comparison
  • DGA entropy scoring
  • Pattern and sequence detection
  • API rule push on detection
analysis: seconds after query
Detection to remediation

From anomaly in the log to
blocked rule in the feed.

When RoCi detects a threat in the telemetry stream, it doesn't just log it. It pushes a rule update back into the synchronous threat intelligence layer — so the next query to that domain is blocked before resolution even begins.

01

Query resolved and logged

A DNS query is authenticated, resolved, and the response returned to the device. Query metadata is written to the telemetry stream asynchronously. The device has its answer. The analysis begins.

response: delivered
latency: 23ms
telemetry: queued for roci
02

RoCi scores the metadata

RoCi ingests the metadata from the telemetry pipeline. Behavioral scoring compares the query against the per-client baseline. Entropy analysis runs on the domain string. Pattern matching checks for known attack sequences across the client's recent query history.

client: eng-0x4f2a
entropy: 4.2 (high)
pattern: probable DGA
03

Threat confirmed — rule update pushed

When scoring exceeds the configured sensitivity threshold, RoCi classifies the threat and immediately pushes a block rule to Nantevo's threat intelligence API. The rule propagates into the synchronous feed lookup used by the live query path.

classification: C2 beaconing
→ block rule: pushed
feed: updated
04

Next query to that domain: blocked

Any subsequent query — from any authenticated client — to the flagged domain now hits a block in the synchronous threat feed lookup and never reaches the resolver. The incident is logged, the affected client is notified, and the detection is visible in the dashboard immediately.

query: xk92mf7...net
result: BLOCKED at feed
incident #4471 logged

This creates a self-reinforcing feedback loop: every detection makes the synchronous threat feed more accurate for every subsequent query across every client. A threat detected on one client's traffic is immediately blocked for all clients — without any manual intervention, without any delay in the query path, and without waiting for a human analyst to review and act.

Detection capabilities

What RoCi looks for
in your DNS traffic.

RoCi's detection capabilities are tuned specifically for DNS-layer threat patterns — the categories of malicious behavior that are invisible to endpoint security but plainly visible in query telemetry.

Domain Generation Algorithm detection

Malware families use Domain Generation Algorithms to programmatically generate large numbers of domain names. The malware tries each in sequence until it finds one that resolves — connecting to the attacker's command and control infrastructure without relying on a hardcoded domain that can be blocked.

DGA domains have a characteristic signature in their structure: high entropy, unusual character distributions, and subdomain patterns that deviate sharply from human-readable domains. RoCi scores every queried domain for these characteristics.

Signal

Subdomain entropy score exceeds 3.5 — character distribution inconsistent with human-readable domain patterns

Signal

Sequential pattern — multiple high-entropy domains queried within a short window, consistent with DGA iteration behavior

Signal

NXDOMAIN cascade — high volume of non-resolving queries to novel domains, indicating the malware is iterating through its generated list

roci — DGA detection // eng-0x4f2a
roci> dga_scan client:eng-0x4f2a

scanning last 60min query window...

domain: api.stripe.com
entropy: 1.1 · classification: benign

domain: cdn.shopify.com
entropy: 1.3 · classification: benign

domain: xk92mf7a3n.cdn-analytics.net
entropy: 4.2 · flagging for review
character distribution: anomalous
known parent domain: no
first seen: this query
classification: probable DGA / C2

→ block rule pushed
→ incident #4471 created

Command and control beaconing

Compromised devices regularly check in with attacker-controlled infrastructure to receive instructions, exfiltrate data, or report status. These check-ins — called beaconing — often use DNS to locate the C2 server, and produce characteristic timing patterns in query logs.

A device making DNS queries to the same novel domain at regular intervals — every 60 seconds, every 5 minutes — is a reliable signal of beaconing behavior. RoCi detects this by comparing query timing distributions against each client's established patterns.

Signal

Periodic queries to the same novel domain at regular intervals — consistent with an automated check-in rather than user-driven browsing

Signal

Off-hours query activity — C2 beaconing continues while the device is idle, producing DNS traffic patterns inconsistent with the user's baseline

Signal

Domain registered recently, low global query volume, hosting on infrastructure associated with bulletproof providers or known malicious ASNs

roci — C2 detection // eng-0x4f2a
roci> beacon_scan client:eng-0x4f2a

analysing query timing patterns...

! periodic pattern detected
domain: update-svc.analyticscdn.net
interval: 300s ±4s (very regular)
first query: 02:14 UTC (off-hours)
query count (8hr): 96
domain age: 3 days
asn: known bulletproof hosting

classification: probable C2 beacon
→ block rule pushed
→ incident #4472 · high severity

DNS tunneling and data exfiltration

DNS tunneling encodes data inside DNS queries themselves — typically using unusually long subdomains or TXT record requests to carry arbitrary payloads. Because DNS traffic is often less scrutinized than HTTP or direct connections, it can be used to exfiltrate data or establish covert communication channels that bypass traditional network controls.

DNS tunneling produces distinctive patterns in query logs: extremely long subdomain strings, high volumes of TXT record requests, and large response payloads inconsistent with normal DNS operation.

Signal

Subdomain strings exceeding 63 characters — DNS allows up to 63 characters per label, and tunneling tools pack data into this space

Signal

Anomalous TXT record query volume — legitimate applications rarely make frequent TXT lookups, making elevated TXT query rates a reliable indicator

Signal

Response payload size anomaly — DNS responses carrying large TXT records or unusually large A/AAAA responses inconsistent with normal resolution

roci — tunnel detection // eng-0x4f2a
roci> tunnel_scan client:eng-0x4f2a

scanning for tunneling indicators...

! anomalous TXT query volume
txt queries (1hr): 847
baseline txt (1hr): 12
deviation: 70x above baseline

! long subdomain detected
dGhpcyBpcyBleGZpbHRyYXRlZCBkYXRh
.exfil-channel.attacker.net
subdomain length: 58 chars (encoded)

classification: DNS tunneling / exfil
→ block rule pushed · domain family
→ incident #4473 · critical severity

Behavioral anomaly detection

Not every threat has a known signature. Behavioral anomaly detection identifies deviations from a client's established baseline that warrant investigation — even when the specific threat type hasn't been seen before.

Because RoCi maintains a per-client baseline rather than a fleet-wide average, it can detect subtle deviations that would be invisible in aggregate data. A device querying 10x its normal volume, suddenly resolving domains in a new geographic region, or generating unusual query type distributions are all detectable signals.

Signal

Query volume spike — significant deviation from the client's per-day and per-hour baseline, which may indicate malware activity or a compromised process

Signal

Novel domain cluster — sudden appearance of many previously-unseen domains outside the client's normal resolution pattern, suggesting new software or infection

Advantage

Per-client baselines mean a device that legitimately queries more — a developer machine, a build server — has a higher normal threshold. No false positives from expected high-volume clients

roci — anomaly detection // eng-0x4f2a
roci> anomaly_scan client:eng-0x4f2a

baseline (30d avg): 2,400 q/day
today so far (6hr): 8,947 queries
projected: ~35,000 today (14.6x)

novel domains (today): 847
novel domains (30d avg/day): 23
novel domain rate: 36x above normal

top new domain cluster:
*.cdn-update-svc.net (412 queries)
domain age: 4 days · low reputation

anomaly score: 9.1/10 · investigate
→ incident #4474 · review required
Per-client baselines

Your normal is not
everyone else's normal.

Fleet-level threat detection averages away the signal. RoCi maintains an independent behavioral baseline for every authenticated client — so anomalies are measured against what that specific device normally does, not what a server farm or endpoint pool looks like.

Client eng-0x4f2a — query volume · 7 days
MonTueWedThuFriSatSun
7-day avg
2,412
queries / day
Today
8,947
anomaly flagged ↑
P99 latency
31ms
7-day trailing
Threats blocked
38
this week
Why per-client matters

Fleet averages hide the signal

A developer workstation that queries 10,000 domains a day is normal for a developer. A corporate laptop doing the same thing on a Tuesday at 3am is not normal for that laptop. Fleet-level baselines cannot distinguish between these. Per-client baselines make that distinction automatically.

Baseline establishment

RoCi learns each client over time

From the moment a client is enrolled, RoCi begins building its baseline — query volume distributions by hour and day, typical domain categories, normal response patterns, and query type ratios. After a short observation window, behavioral scoring becomes active and thresholds are set to that client's specific normal.

Adaptive thresholds

Baselines evolve as behavior changes

If a client's usage pattern legitimately changes — a new role, new software, a new workflow — the baseline adapts over time rather than generating persistent false positives. Sudden deviations still trigger alerts; gradual legitimate evolution updates the model.

Cross-client intelligence

A threat detected on one client protects all

When RoCi pushes a block rule from a detection on one client, that rule propagates into the shared threat feed immediately. Every other authenticated client across the platform benefits from the detection without their individual queries or baselines being affected.

Full capability set

Everything RoCi watches for.

A complete reference of RoCi's detection and analysis capabilities, all operating asynchronously on the telemetry stream without any impact on query response times.

DGA detection

Domain generation algorithm identification

Shannon entropy scoring on subdomain strings, character distribution analysis, and NXDOMAIN cascade detection. Catches malware iterating through algorithmically-generated domain lists in search of live C2 infrastructure.

threshold: entropy > 3.5
method: per-query + sequence analysis
C2 beaconing

Command and control check-in detection

Timing distribution analysis across query sequences to identify regular, automated intervals inconsistent with user-driven DNS traffic. Off-hours activity detection and newly-registered domain correlation.

threshold: interval regularity + domain age
method: timing distribution analysis
DNS tunneling

Data exfiltration via DNS channel

Long subdomain detection, anomalous TXT record query volume analysis, and response payload size monitoring. Identifies covert channels using DNS as a data transport mechanism to bypass traditional network controls.

threshold: subdomain length + TXT volume
method: payload analysis + query type ratio
Behavioral anomaly

Per-client baseline deviation detection

Volume spike detection, novel domain cluster identification, and query pattern deviation scoring — all measured against the specific client's established normal, not a fleet-wide average. No false positives from legitimately high-volume clients.

baseline: rolling 30-day per client
method: z-score deviation scoring
Threat feed enrichment

Automatic rule propagation on detection

When any detection exceeds threshold, RoCi immediately pushes a block rule to Nantevo's threat intelligence API. The rule propagates to the synchronous feed lookup used by all active clients — turning a detection into a cross-platform block in seconds.

propagation: seconds after detection
scope: all authenticated clients
Incident logging

Structured incident records per detection

Every detection generates a structured incident record — client identity, domain, classification, confidence score, evidence signals, and rule action taken. Incidents are immediately visible in the dashboard and available via API for SIEM integration.

format: structured JSON incident record
available: dashboard + API + SIEM
Origin

Born on the road.
Trained on two years
of real traffic.

RoCi's name comes from Rocinante — the van that carried the engineer who built Nantevo across two years and over 40,000 miles of the United States. Throughout that journey, a Raspberry Pi mounted in the dashboard ran continuously, monitoring DNS traffic, logging queries, and building the dataset that became RoCi's foundation.

Two years of production DNS telemetry from real clients across real networks — mobile connections, satellite internet, hotel WiFi, coffee shops, data centers — is an unusually rich and varied dataset for a threat intelligence model. RoCi knows what normal looks like across a wide range of network conditions because she has seen it.

"RoCi doesn't look for threats. She looks for things that don't look like everything else she's seen before."

2023 — Origin

Raspberry Pi in a ProMaster van

The first version of RoCi was a Raspberry Pi mounted in the van's dashboard — monitoring DNS traffic across mobile networks, satellite connections, and public WiFi during a two-year nomadic journey across the US.

2023–2024 — Data foundation

Two years of production telemetry

Continuous collection of DNS query metadata across real networks and real clients. Not a synthetic dataset or a lab environment — actual traffic from production deployments across a wide range of network conditions and use patterns.

2024 — Integration

RoCi moves into the Nantevo platform

The intelligence model built on van-era telemetry is integrated into Nantevo's asynchronous analysis pipeline. Per-client behavioral baselines, DGA scoring, and the detection-to-remediation loop become production capabilities.

Now — Continuous learning

Every detection makes RoCi sharper

Each incident confirmed by an analyst, each rule update pushed, each false positive corrected refines RoCi's model. The platform's telemetry is the training data. Every client enrolled extends the dataset.

See RoCi detect a threat
in a live environment.

Demo includes the full detection pipeline — from enrollment through behavioral baseline establishment to a live anomaly scenario.