Deep Dive · The Endpoint Agent

One agent. Supported where you operate.

The Edge agent is packaged for Windows x86-64, Linux x86-64 and Apple Silicon macOS. It collects native telemetry, runs deterministic detection and, on Windows, can isolate a host when a high-confidence ransomware policy fires. Other host isolation remains approval-gated. The agent can buffer while disconnected; a fully self-hosted, air-gapped platform remains on the roadmap.

Platform Coverage

Built for segmented and intermittently connected estates.

The agent uses a common codebase with platform-specific collectors and packaging. Supported release channels are explicit: Linux x86-64, Windows x86-64 and Apple Silicon macOS. Package support and collection depth still depend on the operating system and host configuration.

Windows

Windows x86-64

Native Windows Event Log subscription with bookmarks and resubscription, covering the Security, System, Application, PowerShell and Sysmon channels. Delivered through a customer-scoped MSI.

Linux

Linux x86-64

Audit-based process monitoring, with eBPF used instead where auditd already owns the audit socket, plus the auth and syslog files where the host writes them. Delivered as customer-scoped DEB and RPM packages.

macOS

Apple Silicon

File-integrity monitoring, software inventory and posture checks, delivered as a customer-scoped notarized package. No configuration profile is needed; process monitoring is not enabled on macOS.

Remote & Segmented

Connection-resilient

Supported hosts continue deterministic evaluation and buffer telemetry when the cloud link is interrupted, then resume transport after reconnection.

Where this matters

Remote sites, manufacturing networks, maritime operations and other segmented estates cannot assume a continuous cloud path. Edg3 keeps deterministic controls on the host and returns buffered evidence to the tenant when connectivity resumes.

Autonomous Ransomware Containment

The one response the agent will run on its own.

Ransomware is the only response class enabled for autonomous isolation by default. On Windows, a high-confidence ransomware verdict can move the host into its platform-specific isolation state without a cloud round trip. A slightly less confident verdict waits up to 8 seconds for a central decision, then the agent isolates locally; other host isolation remains approval-gated.

01

Detection fires

Endpoint evidence is evaluated against the active deterministic rules and the configured autonomous-isolation confidence threshold.

02

Agent evaluates locally

A high-confidence ransomware verdict can trigger the local isolation path. A marginal signal is held for the bounded central-decision window.

03

Host isolation

The platform-specific firewall path restricts host connectivity while retaining the Edg3 control path needed for review and recovery. On Windows, connections opened before isolation are not yet cut.

04

State is recorded

Isolation transitions are written to the local state and published for the tenant investigation timeline, waiting in a bounded in-memory buffer while the link is down.

05

Safety and human review

The analyst can confirm or clear the action. A safety deadman is on by default: it lifts any isolation after 30 minutes without the control channel, and after 24 hours even on a healthy link unless the console is renewing an emergency isolation.

Why not everything?

Autonomous containment beyond ransomware is a shorter-term cost saving and a longer-term incident. An agent that quarantines the wrong account at 03:00 is the kind of story that ends product evaluations. Edg3 therefore keeps autonomous containment narrow and routes every other isolation through explicit approval.

Gated Response

Human-approved isolation. Platform-audited.

Outside autonomous ransomware containment on Windows, the platform dispatches two response actions: host isolation and forensic collection. Host isolation never fires without a recorded analyst approval, and the approver’s identity stays attached to the workflow record. Forensic collection runs automatically on Windows when a critical detection matches certain attack techniques, and runs when an analyst approves a host isolation. Only a console Admin can start one by hand, and a memory capture of the Windows credential store (LSASS) is never collected automatically. From agent v0.4.80, the launch build, a vulnerability finding never isolates a host, and the older unsigned command lane is retired: the only command path to an endpoint agent is the Ed25519-signed forensic channel.

Action Default mode Typical approver Audit artefact
Ransomware containment (Windows) AUTO · (platform) Isolation state + response event
Host isolation HUMAN SOC tier-two Approval + operator identity
Forensic collection on a critical detection (Windows) AUTO · (platform) Command audit + triggering detection
Forensic collection by hand HUMAN Console Admin Command audit + operator identity

These are the response actions the platform dispatches today. Other remediation stays with your existing teams and tools.

Playbooks as Code

YAML, version control and explicit approvals.

Edg3 orchestration flows are YAML and version-controlled. Changes can be reviewed as code, and the flow definition exposes the trigger, approval gate and downstream action without a separate proprietary workflow representation.

# Illustrative shape; the deployed schema is flow-specific. trigger: validated_detection approval: required: true scope: tenant action: type: configured_provider_action record: - operator_identity - decision - provider_result

Slack approval and notification flows are present in the orchestration layer, alongside the host isolation and forensic collection actions.

Deployment

Guided deployment with customer-scoped packages.

Infrastructure as code

The current provisioning workflow creates a dedicated AWS-backed tenant instance and its supporting services. Additional hosting models should be treated as roadmap until tested.

Agent rollout

Customer-scoped Windows MSI, Linux DEB/RPM and macOS package workflows produce the installer paths used during onboarding.

Air-gapped install Roadmap · Q4 2026

A full offline installer with bundled models, signed images and update channel is on the roadmap — update cadence controlled by your SCIF policy, not our release train.

Sovereign deployment Roadmap · Q4 2026

A customer-operated control plane in a chosen environment, with its own release and update channel, is on the roadmap.

Agent Security Properties

The agent that watches you is itself worth auditing.

A security agent with a large privilege surface is itself a high-value target. Edge is architected on the assumption that the next supply-chain attack will target a vendor like us. The mitigations below are the baseline, not the premium tier.

Your red team is welcome

Security evaluation should include package provenance, tenant messaging scope, update rejection, offline behaviour and isolation recovery. Ask the product team for the current validation evidence for the release you intend to deploy.