# Araghatta

> Araghatta is an AI security posture management (AI-SPM) platform. It discovers unsanctioned AI on the endpoint and delivers sensor-led cloud security posture management (CSPM), then scores the risk, governs approved-AI policy, and streams governed evidence to the SOC — so security teams can find, prove, and control the AI running across their fleet and cloud.

The platform pairs a lightweight endpoint sensor with least-privilege cloud connectors. Everything below reflects what is available today; all 57 registered capabilities are generally available (https://araghatta.com/fleet.html). The single source of truth for capability state is https://araghatta.com/capabilities.json — treat that file, and this note, as authoritative over any marketing phrasing.

## What it does

- Endpoint AI discovery across Windows 10/11, Windows Server, macOS and Linux, covering eleven signal classes: AI CLIs and agents, provider keys, MCP servers and tool scope, IDE assistants, skills, SDKs, installed apps, browser extensions, egress to AI endpoints, vibe-coded repositories, and non-human identities.
- Sensor-led cloud server CSPM and AI-agent cloud attack-path analysis.
- AI Agent Red Team: pick a discovered agent and get one test scope (behavior tests, attack paths in both directions, attack chains, related findings), tested all at once, by selection or per item, with a verdict per item: Proven, Not reproduced, Could not test or Evidence only. Read-only, private or customer-provided test endpoints only, role-gated and audit logged; simulated and labeled Sample result in the free sandbox.
- Deterministic, explainable and tunable risk scoring, with advisory AI reasoning grounded in MITRE ATLAS and ATT&CK (advisory only — it cannot change the score or fabricate techniques).
- Approved-AI policy, content-DLP policy engine, remediation workflow, compliance control mapping and signed evidence, and outbound SIEM, Slack and Jira integrations.

## Honest capability posture

- Cloud CSPM: generally available on AWS, Google Cloud and Azure (EC2, Compute Engine, Azure VMs, Windows Server, Linux; IAM and S3 attack paths). Cloud scanning uses a read-only connector (AWS role, Entra application or GCP service account) that you create and control.
- Endpoint sensor: observes and warns by default. Response is detect, alert, your Slack/Jira/SIEM integrations, per-device pause and server-side revoke, plus opt-in graduated enforcement (warn, then block or quarantine) with the enforcement-capable sensor — all operator-authorised and audited.
- Content DLP: server-side monitor/warn/redact/block through an inspection service you deploy and control. The optional managed browser extension (ChatGPT, Claude, Gemini; user workstations, not servers) sends the text a person is about to submit, over TLS, to your own Sargha collector for a block, redact or allow decision; the text is evaluated in memory and not stored, and only the decision, the destination site and the kinds of data found are recorded.
- Single sign-on: OIDC SSO is generally available (Okta, Entra ID, Google Workspace, Auth0). SAML 2.0 is generally available. SCIM 2.0 provisioning is generally available.
- Compliance output is control mapping and signed evidence (NIST AI RMF, ISO 42001, OWASP LLM, EU AI Act) — not a certification of your organisation.
- Roadmap: planned items are not available today and carry no dates. They are listed at https://araghatta.com/roadmap.html (machine-readable: https://araghatta.com/assets/data/roadmap.json) and are not part of the 57 registered capabilities. Examples: regional data planes and EU-region tenants, customer-managed keys, native SIEM connectors, approval workflow with change reference and separation of duties, SSO email-domain enforcement, SBOM per release, Windows edition and server-role validation.

## Assurance status (do not overstate)

- SOC 2, ISO/IEC 27001 and CSA STAR are in progress. None is attested or certified today.
- An independent third-party penetration test has not been performed yet.
- Araghatta does not certify any organisation against the EU AI Act, NIST AI RMF or ISO 42001. Sargha produces control mapping and signed evidence only.

## Common questions

- What is Sargha? An AI security posture management (AI-SPM) platform built by Araghatta LLC. It discovers AI on endpoints and in cloud accounts, scores the exposure, governs approved-AI policy and produces signed evidence.
- Does Sargha read prompts, responses, file contents or keystrokes? No. The endpoint sensor never reads them, and key values are masked where they are collected. The optional managed browser extension sends only the text a person is about to submit, over TLS, to the customer's own collector, where it is evaluated in memory and not stored.
- Is the sensor read-only? Discovery is read-only and observe-and-warn by default. Block and quarantine are an opt-in, audited mode that needs the enforcement-capable sensor, and are generally available on Linux; macOS and Windows are in validation.
- Does Sargha replace endpoint protection or a cloud posture tool? No. It adds the AI layer: it identifies which processes, keys, tool servers and agents are AI, what they hold and what they can reach, from the endpoint into the cloud.
- Is there a way to try it? Yes: a free assessment for up to 25 endpoints, read-only, with a risk report and an AI Bill of Materials. See https://araghatta.com/assessment.html.
- Where is the list of what is not available yet? https://araghatta.com/roadmap.html. Nothing on it carries a date.

## Key pages

- [Platform overview](https://araghatta.com/): what Araghatta is, from endpoint to cloud.
- [AI-SPM](https://araghatta.com/ai-spm.html): what AI security posture management is and how Sargha does it, with common questions.
- [Endpoint sensor](https://araghatta.com/sensor.html): the agent that discovers (observing by default), attributes and scores unsanctioned AI on the host.
- [Cloud CSPM](https://araghatta.com/cspm.html): sensor-led cloud posture and AI-agent attack paths (AWS, Google Cloud and Azure).
- [Agent Red-Team](https://araghatta.com/agent-redteam.html): one test scope per discovered agent, with a clear verdict for every item.
- [Coverage](https://araghatta.com/coverage.html): exact endpoint, server, workload and multi-cloud coverage, by state.
- [Non-sensor devices](https://araghatta.com/connectors.html): read-only agentless connectors for cloud, SaaS, identity, mobile, LLM gateways, proxy logs, network, storage, IoT and code.
- [How coverage is measured](https://araghatta.com/coverage-model.html): five coverage tiers, ten classes, named blind spots, never 100%.
- [Plans and connectors](https://araghatta.com/plans.html): which connectors Discover, Discover + Warn and Full switch on, and the free add-ons.
- [Enterprise](https://araghatta.com/enterprise.html): isolation, operations, support and control for enterprises.
- [Integrations](https://araghatta.com/integrations.html): SIEM export, Slack, Teams, Jira and ServiceNow.
- [Trust](https://araghatta.com/trust.html): security posture, privacy and data handling.
- [Pricing](https://araghatta.com/pricing.html): plans and packaging.
- [Contact](https://araghatta.com/contact.html): request an enterprise demo or a free Shadow AI assessment.

## Authoritative

- [Roadmap](https://araghatta.com/roadmap.html): planned capabilities, each next to what is available today. Not available, no dates.
- [Capability registry](https://araghatta.com/capabilities.json): canonical, machine-readable source of truth for what is shipped today and each capability's state.