console v1.70.151
0/0agents online
0events/sec
0events seen
0open alerts
connecting
Endpoints
Endpoints
No agents yet. Start the AI EndSight agent on a Windows host pointed at this server.
Live telemetry
Endpoints
Telemetry all · none
Group by
Live telemetry is paused.
Detections
Detections
No detections yet. Try the test playbook in the README.
Response actions
Kill, quarantine and isolate actions appear here with their results.
Overview
Executive Insight
Ransomware
Alerts
Alerts
Dots Universe Cyber Mosaic: correlating the Dots
Alerts
contains
Search
Investigate/Search

Search

Query every stored event like Elastic or Kibana: fields, AND, OR, NOT, wildcards, contains, regular expressions, ranges.
Insert
Incidents

Endpoint risk

top incident + 15% of others, 24 h half-life

Open incidents by risk

ATT&CK tactics in open incidents

AI verdicts

open incidents and closed by AI in the period

AI confidence

incidents the AI analysed

Incident status

AI isolation

Smart Isolation

Response times

last 30 days
RiskIncidentEndpointStatusVerdictAI verdictAI confidenceAI isolationDetectionsKill chainStorylinesUnmitigatedLast activityAssignedAction

Closed automatically by AI

RiskIncidentEndpointClosed by AIConfidenceWhy the AI closed it
Detection validation
Evasion kit
Click a technique for its rules, atomic tests and results. Windows techniques from MITRE ATT&CK® (© The MITRE Corporation); tests from Atomic Red Team (MIT licence).
Test runs
StartedEndpointAtomic testResultAlertsTelemetry
IdentitySight
AI Security
Storylines Behaviour engine
Campaigns
Forensics
ExposureSight
Dark Web Monitoring
Vulnerabilities
Remote Support
Network EndSight
Loading
Assets
Endpoints
EndpointStateUsersPolicy and groupPublic IPPrivate IP address(es)AgentHealthScanLast seenTags
Policies

Policies

what agents collect, how they respond, what they ignore
PolicyModeProtects againstSensorsTamperUsed by groupsEndpointsUpdated

Endpoint groups

an endpoint gets the policy of its highest-priority (lowest number) group, else Default
PriorityGroupMembershipPolicyMatchingEffective
Administration

Console users

accounts and roles
UserRoleEmailTwo-stepStateLast sign-in
TimeUserAddressActionOnStatusDetail

Every change made through the console or API is recorded (who, from where, what, result), plus sign-ins, denied attempts, agent enrollments and command results. Secrets are masked.

Container security
Reporting
Agent repository

Scope tokenconsole address, certificate and enrollment token in one value

...

Platformsclick to filter

PackageVersionTypeArchSizePublishedSHA-256

The console keeps the current agent and the previous five. Setup programs and MSI packages downloaded here already carry this console's address and certificate; With token also builds in the scope token, so they install and enroll with no typing (keep those files as private as the token). Linux packages are installed with the copied command, which downloads the chosen version and checks its SHA-256. Regenerating the scope token never disconnects enrolled agents.

DataSight
Protect/DataSight

Overview

File Vault

In the vault

kept encrypted

Verdicts

click to filter

Analysis

where files are analysed

Malware families

named by the sandbox
Quarantine

In quarantine

files that cannot run

Threat categories

click to filter

Malware families and tools

by threat name

Endpoints

most files in quarantine

Found by

what took the file

Quarantined per day

last 14 days

Recovered

released from quarantine, last 30 days

Outcome

every file quarantined, last 365 days

On several endpoints

the same file in more than one quarantine

Quarantined files are encrypted on their endpoint and cannot run. Release puts a file back where it was. Release and allow also decides for its SHA-256 on every endpoint at once: the pre-execution check lets it run, automatic responses leave it alone, and the machine learning model learns it is clean. Block stops the file on every endpoint (or on the endpoints of chosen policies) and can sweep the endpoints where it was seen to quarantine the copies still there. Agents report their quarantine whenever it changes; Refresh asks the others.

Scans

Quick: running programs and their memory, what starts at sign-in or boot, and the folders malware lands in (a few minutes). Full: every local drive, the files that can carry code. Custom: the files, folders or drives you name, every file in them. Each file is checked cheapest first: unchanged since it was found clean, analysts' decisions and the blocklist, code signature, threat rules, file analysis, then the console (threat intelligence, the fleet, reputation). Scans run at background priority, use at most the policy's share of one processor core and wait while the computer is busy or on battery; scans users start from Explorer's right-click menu run faster. In Protect mode malicious files are quarantined; everything found is an alert (EDR-049, EDR-050 for memory, EDR-051 for Microsoft Defender).

Threat rules

Rule sourcesturned-on sources are compiled into one rule set and sent to every agent

OnSourceConfidenceRulesUpdatedStatus

Your rule fileswritten here; checked when saved

Rules in use

RuleSourceConfidenceDescriptionAuthor

Threat rules describe known malware, hacking tools, web shells and malicious documents by their content. Agents use the rules in scans, when a file is written to disk (On write), before a file from outside opens and in the memory of running programs. Confidence: "From each rule" reads the rule's severity or score (high and critical are malicious and quarantined in Protect mode; medium and low are suspicious); "Suspicious only" reports every match for an analyst and never acts on it, a safe way to try a new feed. Feeds are downloaded once a day by this console; a file that does not compile is left out and shown here.

Machine learning
Out of shadow mode, with guardrails (1.70.54). Agents score programs when they are written, before they open and when an unknown one starts; borderline files get a second opinion from the console's larger model (from the file's numbers, never the file, so it works air-gapped too). Each file gets one verdict that also weighs its code signer, how common it is across your endpoints, reputation, sandbox, origin and behaviour. A verdict of malicious at 97% or more that something besides the classifiers backs (reputation, threat intelligence, the sandbox, a malicious behaviour, an invalid signature) blocks the file on every endpoint; never the classifier alone, never a file signed by a known publisher. A new model goes live by itself only when it beats the one in use on this console's labelled files.

Intelligence servicelearning across CyberArm's clients: opt-in, numbers only, shared once 3 organisations saw the same thing

Learned modelstrained on this console from analyst and AI verdicts, the File Vault and the sandbox; used only when they rank held-back verdicts better than the rules

Decision stageevery alert judged on the evidence the console holds, before an analyst sees it

Known activityalerts about normal activity of known programs closed as benign, with the reasons (catalog, analysts' verdicts, fleet baseline, machine learning)

How a verdict is formedevidence each signal adds, in log-odds

Agreement with reputation servicesfiles both know, at the "malicious" cut-off

Training data from your endpointsfeature vectors (numbers, never files) labeled from strong evidence, for the next models

Collect Download

Scored files

FileEndpoint scoreSecond opinionVerdictSignerHostsLast seen

Modelsendpoint models run on agents; console models give the second opinion

RoleModelTrainedTreesSizeTest AUCStatus
Threat intelligence

LookupIP, domain, URL or file hash

Your indicatorsanalyst IOCs, matched like feeds

Indicators Severity Description Added indicators are retro-hunted over the last 24 h.

Feedsrefreshed automatically; cached for offline restarts

OnFeedTypeSeverityIndicatorsUpdatedStatus

Fleet prevalenceexecutables and drivers seen across endpoints

BinarySHA-256HostsRunsFirst seenSignature

Feed credentials and response

abuse.ch Auth-Key Enables MalwareBazaar (hashes), ThreatFox and URLhaus. abuse.ch states commercial use may need a paid subscription. MISP URL MISP API key Known malware
Rules

Detection rules

what raises alerts

ATT&CK coverageenabled rules per tactic

OnRuleSourceLevelLog sourceATT&CKModeHits 24hNoiseStatus

Fleet noise

the same rule firing on the same program on many endpoints: one item, decided once for the fleet and for the alerts already in the queue

Suppression rules

exemptions: pushed to every agent and enforced on the endpoint (no automatic kill or quarantine, offline too); matching alerts never reach the alert queue
OnSuppression ruleConditions (all must match)SuppressedHiddenStateCreated

Notification rules

email the right people when alerts, incidents, ExposureSight findings or agent health match; proximity counts repeats so a burst sends one email and a summary
Loading...
Overview

Suspicious activities

Time / hostSeverityRule / ATT&CKProcessPath / targetSummaryStatusActions
AI EndSight by CyberArmAI EndSight by CyberArm

AI EndSight Control Center

Sign in to continue

© CyberArmNext-generation endpoint detection and response
Endpoint detection and responseSee every layer.
Find every threat.
From the network edge down to the kernel, EndSight looks deeper into every endpoint and stops what hides there.
Welcome back
VERIFYING CREDENTIALS