Platform scope
Clients, LTSC and server roles
- Windows 10 and 11 Enterprise, Pro and LTSC
- Windows Server 2016, 2019, 2022 and 2025
- Server Core, domain controllers, RDS and failover clusters
- x64 and ARM64 release channels
Enterprise coverage matrix
Coverage without hand-waving.
See exactly where Araghatta runs, what it observes, how it is deployed and which actions are available—from Windows endpoints and domain controllers to Kubernetes workloads and cloud AI services.
Windows & serversWindows 10/11, Server 2016–2025, Server Core, RDS and domain controllers.
Linux & macOSSigned native sensors, scheduled reconciliation and real-time eBPF signals on Linux.
VDI & workloadsAVD, Windows 365, Citrix, Horizon, Kubernetes, containers and CI runners — on the roadmap.
Cloud connectorsAWS available today; Azure and Google Cloud via read-only connector (on connect).
Supported environments
One control plane, explicit collection models.
Endpoint, server, virtual desktop, workload and cloud resources are not treated as identical assets. Each has its own identity, telemetry, deployment and response contract.
| Environment | Status | Deployment | Discovery model | Response model |
|---|---|---|---|---|
| Windows 10 & 11 | GA | Intune, Configuration Manager, GPO, MSI or scripted install | Continuous event-assisted detection plus scheduled inventory reconciliation | Observe-only sensor; pause and server-side revoke today; warn/redact/block/quarantine via the customer-authorised policy path (roadmap) |
| Windows Server 2016–2025 | Beta | Configuration Manager, GPO, PowerShell, image or configuration management | Service, process, package, identity, network and configuration signals | Observe-only; pause and revoke today; server-scoped policy enforcement on the roadmap |
| macOS | GA | Jamf, Kandji, Intune or signed package | Native endpoint telemetry and scheduled reconciliation | Observe-only; pause and revoke today; user-notification and content policy on the roadmap |
| Enterprise Linux | GA | RPM/DEB, Ansible, systemd, golden image or configuration management | eBPF exec/connect telemetry plus scheduled package and configuration inventory | Observe-only; pause and revoke today; opt-in host/process deny-list (advanced, off by default); AI-aware enforcement on the roadmap |
| VDI & multi-user | Roadmap | Golden image, non-persistent clone or session-host deployment | Stable machine template plus session/user attribution and duplicate suppression | Collection model in development; user- and pool-scoped policy on the roadmap |
| Containers & Kubernetes | Roadmap | DaemonSet, host sensor, image scan and API connector | Image AI-BOM, runtime process/egress, workload identity and MCP endpoints | Collection model in development; admission/policy integration on the roadmap |
| Cloud control planes | AWS GA | Least-privilege organisation connector using customer-controlled identity (AWS today; Azure & Google Cloud on connect) | API inventory, event stream and security-log correlation | Alert, ticket and revoke today; remediation through customer-approved runbooks |
Windows & Windows Server
Built for managed fleets and critical servers.
The Windows sensor uses a native service with no general-purpose runtime dependency, verified by SHA-256 checksum (Authenticode signing on certificate procurement). The Windows client sensor is GA and the Windows Server sensor ships today as a beta; the collection model below describes what each reports. Client, server, VDI and privileged-host behavior are documented separately.
Deployment & lifecycle
Use the tools IT already operates
- Intune, Configuration Manager/SCCM and GPO
- MSI and PowerShell silent deployment
- Golden-image support for AVD, Citrix and Horizon
- Phased update rings, pinning, rollback and uninstall controls
Runtime safety
Predictable endpoint behavior
- Dedicated Windows service and least-required privileges
- No reboot for install or routine upgrade
- Proxy/PAC and TLS-inspection compatibility
- Health telemetry, bounded spool and redacted support bundles
| Windows signal | Source | Data reported | Content boundary |
|---|---|---|---|
| AI applications & CLIs | Process, service, installed package and signer metadata | Name, version, publisher, hash, execution state, machine and accountable identity | No executable content or source code |
| MCP servers & agents | Process and approved configuration locations | Package, version, provenance, tool scope and runtime state | Configuration values are redacted; credentials are never transmitted |
| Keys & secrets | Environment and bounded configuration discovery | Provider, masked fingerprint, location class and exposure context | Usable secret values never leave the device |
| IDE & browser extensions | Marketplace identifiers and extension metadata | Stable ID, name, version, publisher and signed-in account class | No browser history, page content, cookies or prompts |
| Network & AI egress | DNS and connection metadata | Destination hostname/IP, port, provider, process and policy decision | No packet payload; URL paths are excluded by default |
| Host and identity context | Operating-system and device-management metadata | Device ID, OS build, domain/tenant, posture, user/service identity class | Identity display can be pseudonymised by tenant policy |
Eleven discovery classes
The same taxonomy across every environment.
Collection technique and latency vary by platform, but findings map to one normalized AI asset, identity, risk and evidence model.
| Signal class | Endpoint & server | Cloud & workload | Evidence produced |
|---|---|---|---|
| Network & egress | Process-attributed DNS/SNI and connection metadata | Flow, audit and gateway-event correlation | Provider, destination, source identity, action and time |
| MCP servers & tool scope | Local runtime and configuration inventory | Hosted endpoints, images and service configuration | Package, version, provenance and granted capabilities |
| API keys & secrets | Masked, on-device fingerprinting | Secret-manager and IAM metadata correlation | Provider, fingerprint, location class and rotation state |
| AI CLIs, apps & SDKs | Process, package and filesystem metadata | Image, function and workload package inventory | Product, version, execution state and owner |
| IDE AI assistants | Stable marketplace IDs and account class | Developer-platform and repository connector context | Extension, publisher, version, user and policy state |
| Agent skills | Bounded metadata and source provenance | Repository, image and build-pipeline context | Skill identity, source trust and privileged capabilities |
| AI-authored repositories | Project footprint and tooling metadata | SCM and CI/CD correlation | Repository, owner, AI tooling and review status |
| Autonomous agents | Runtime, parent process and unattended state | Service, job, function and orchestration events | Agent, model, identity, tools, credentials and autonomy |
| Browser extensions | Stable extension ID inventory | Managed-browser policy correlation | Extension, publisher, version, user and sanction state |
| Mobile AI apps | UEM/MDM application inventory | Identity and sanctioned-app correlation | Application, version, account class and device posture |
| Cloud AI posture | Instance identity and metadata-service posture | Organisation-wide AI services, IAM and data-path inventory | Resource, identity, policy, region, exposure and owner |
Servers, VDI & workloads
Identity that survives modern infrastructure.
Persistent servers, non-persistent desktops and ephemeral workloads use different deduplication, retention and ownership rules.
Servers & VMs
Physical, virtual and cloud-hosted
Windows and Linux servers, VMware/Hyper-V guests, cloud VMs, bastions and developer hosts with service-account attribution.
Virtual desktops · roadmap
Persistent and non-persistent
AVD, Windows 365, Citrix and Horizon with golden-image deployment, clone identity and interactive-session attribution — collection model in development.
Kubernetes · roadmap
Node, image and runtime views
EKS, AKS and GKE coverage through DaemonSet, image scan and API connector, with namespace and workload identity — on the roadmap.
CI/CD & serverless
Build and ephemeral execution
GitHub, GitLab, Azure DevOps and Jenkins runners plus Lambda, Functions, Cloud Run and Cloud Functions.
AWS today · Microsoft Azure & Google Cloud on connect
Multi-cloud AI posture, not just host context.
AWS is generally available today. The Azure and Google Cloud connectors are read-only and being brought online — the models below describe what each connector collects on connect. Self-contained workload sensors combine AI runtime and host evidence with least-privilege, read-only organisation connectors for identity, policy, storage and control-plane posture. Neither path requires Araghatta to retain permanent cloud credentials.
Cloud-server CSPM joins runtime to configuration.
An exposed metadata-service configuration is not reported as an isolated checkbox. It is correlated with the AI agent actually running on the server, the workload identity it could act through, effective permissions and reachable cloud data services. The result includes preconditions, confidence, blast radius and the control that breaks the path.
Amazon Web Services · GA
Organisation-wide AWS visibility
- Organizations, accounts, OUs, regions and Control Tower
- Bedrock, SageMaker, Amazon Q, EC2, ECS/EKS and Lambda
- IAM, roles, SCPs, workload identity and AI-related data paths
- CloudTrail, Security Hub, GuardDuty, Config and Security Lake/OCSF
- Cross-account role, External ID, CloudFormation/Terraform onboarding
Microsoft Cloud · On connect
Azure, Entra and Microsoft AI
- Tenants, management groups, subscriptions and regions
- Azure OpenAI, AI Foundry, Machine Learning, VMs, AKS and Functions
- Managed identities, service principals, OAuth grants and Conditional Access
- Defender XDR, Defender for Cloud, Sentinel and Activity Logs
- Enterprise application, Azure RBAC and Private Link onboarding
Google Cloud · On connect
Google Cloud and Gemini estate
- Organizations, folders, projects, regions and billing context
- Vertex AI, Gemini, Model Garden, GKE, Cloud Run and Functions
- Service accounts, IAM bindings, OAuth clients and workload identity
- Security Command Center, Audit Logs, VPC Flow Logs and Google SecOps
- Workload identity federation and service-account onboarding
| Connector contract | Enterprise behavior | Customer control |
|---|---|---|
| Access | Read-only, least-privilege identity with exact published role/scopes | Customer creates, restricts, rotates and removes the identity |
| Collection | Event-driven where supported, API polling for reconciliation, host/runtime telemetry for execution context | Scope by organisation, account/subscription/project, region and service |
| Content | Resource, identity, policy, configuration and event metadata | Prompts, responses, object bodies, secret values and source code excluded by default |
| Actions | Alert, ticket and revoke today; quarantine and policy remediation on the roadmap | Dry-run, approval, scoped service role, audit, rollback and emergency disable |
| Private operation | PrivateLink/Private Service Connect, customer-managed keys and customer-cloud deployment | Region, network path, key ownership, retention and operator access policy |
Validate coverage against your estate.
Bring your Windows, server, VDI, Kubernetes and cloud inventory. We will map deployment, permissions, data fields, integrations and response controls before any sensor or connector is enabled.