Data sources
Data sources and operating surfaces
The product should support three honest modes: launch direct scans, connect to a read-only source, or ingest evidence that the customer already centralizes elsewhere. This page makes those boundaries explicit and also shows where runtime, fleet, graph, and policy surfaces fit once the data is in the system.
Direct scans
Agentless or local scan jobs that agent-bom launches directly.
Local inventory and MCP discovery
Scan local MCP configs, Python agent projects, GitHub Actions, inventories, container images, and Terraform from the New Scan flow.
Kubernetes and image analysis
Use the same scan flow for cluster inventory and container/image package analysis where you can point agent-bom at the runtime or artifact directly.
Connected sources
Read-only sources where the customer points us at a cloud or SaaS system that already contains the data.
Governance and cloud activity
Snowflake-backed governance, access history, and activity pages already consume cloud-side telemetry without forcing everything through the local scan form.
Connector-backed discovery
The backend exposes connector and SIEM connector routes today. The product still needs a first-class setup wizard in the UI.
Ingested evidence
Evidence pushed into agent-bom from an existing collector, exporter, or security data lake workflow.
OTLP traces and runtime events
The traces surface and POST /v1/traces route accept OTLP JSON so teams can correlate runtime calls against known vulnerable assets.
Security lake and warehouse feeds
If customers already centralize evidence in Snowflake or another data platform, agent-bom should consume that source of truth instead of duplicating collection.
Imported artifacts
Customer-exported files that agent-bom can analyze without managing the source system.
Operating surfaces after ingest
Discovery and ingest are only the front door. Agent-bom also needs clear surfaces for runtime review, fleet operations, policy enforcement, and graph analysis after the data lands.
Security graph and path analysis
Persisted graph snapshots, attack-path focus, and blast-radius analysis across agents, servers, packages, tools, and credentials.
Fleet management
Persisted agent inventory, lifecycle state, trust score, and operational review state for the agent fleet.
Runtime proxy and alerts
Live runtime enforcement, detector alerts, drift protection, and audit review for MCP and tool-call activity.
Gateway and policy enforcement
Policy evaluation, review, and enforcement surfaces for controlling high-impact tool usage and approval workflows.
Guardrail principle
Prefer agentless read-only discovery when the product can safely gather the data itself. When the customer already owns the collection path, use imported artifacts or pushed ingest instead of rebuilding their telemetry pipeline inside agent-bom.