SightHelm

What it does

• Discover with evidence. Read-only collectors surface agents and record freshness, confidence, limitations and failures. Missing or stale evidence becomes a visible coverage gap, never a clean assumption.
• Map reach and responsibility. A content-free graph connects agents to identities, permissions, owners, resources, tools, MCP servers, users, departments and governed data domains.
• Evaluate deterministic policy. A bounded policy-as-code engine uses independently approved, immutable versions and keeps an LLM out of the decision path.
• Make accountable decisions. Business sponsors and technical owners carry explicit responsibility, while exceptions require justification, expiry and maker-checker approval.
• Govern and enforce action. Remediation plans move through approval, dispatch and execution. Separately enabled worker credentials can disable a supported Microsoft Entra service principal when policy pre-authorises containment.
• Prove the outcome. Append-only activity, digest-protected exports and reports, Azure Monitor and Sentinel delivery, and signed event integration preserve attributable evidence.
• Separate hosting from coverage. Run one customer-controlled appliance in the approved hosting location, then connect independently authorised provider estates without requiring the host cloud and source cloud to match.

• Console and control API. One operational surface for inventory, agent detail, coverage, policy, exceptions, remediation, evidence, deployment and governed commands.
• Provider connector packs. Source-specific collectors retain provenance while the correlation engine resolves a canonical inventory and materialises relationships across provider boundaries.
• Risk, policy and governance. Deterministic scoring and policy evaluation feed accountability, disposition, recertification, exception and notification lifecycles.
• Remediation and enforcement. Durable workers run maker-checker plans, external runbooks and the narrowly scoped, opt-in containment actions SightHelm supports.
• Evidence and integrations. PostgreSQL holds inventory, immutable history and transactional queues; evidence can flow to reports, exports, Sentinel, Azure Monitor and allowlisted signed endpoints.
• Customer-controlled appliance. The signed OCI image and Helm package target AKS, EKS, GKE, OpenShift and conformant private Kubernetes through explicit database, registry, secret, ingress, OIDC and telemetry contracts. AKS is the currently accepted reference profile.

• Microsoft — implemented baseline. Entra Agent ID and Graph, Defender/Agent 365, Power Platform and Copilot Studio, Microsoft Foundry, Purview and Application Insights.
• Salesforce — inventory baseline. Agentforce inventory is available now; identity, permissions, actions, data reach and tracing remain gated follow-on layers.
• Google Cloud — planned connector pack. Vertex AI Agent Engine with service-account, IAM, Cloud Asset Inventory and content-free audit evidence.
• AWS — planned connector pack. Amazon Bedrock AgentCore and Agent Registry with IAM, CloudTrail, CloudWatch and resource metadata.
• Hosting — portable by contract. The host platform supplies compute, PostgreSQL, image registry, secrets, ingress, OIDC and operations; provider collectors receive separate worker-only credentials.