Skip to content

Infrastructure knowledge graph

DXDT continuously maps your cloud infrastructure into a living dependency graph — every resource, every relationship — and uses it to answer topology, blast-radius, and drift questions in chat and in the graph explorer.

What is mapped

Nodes are typed as: Service, Compute, Database, Cache, Queue, Storage, LoadBalancer, Network, DNS, CDN, Region, CloudProvider. Each node carries its cloud (AWS, GCP, Azure, on-prem), region, and provider metadata such as instance type and tags.

Relationships between resources are typed as: DEPENDS_ON, RUNS_ON, STORES_DATA_IN, CACHES_WITH, PUBLISHES_TO, CONSUMES_FROM, EXPOSED_VIA, IN_SUBNET, IN_REGION, IN_CLOUD, PEERS_WITH, RESOLVES_TO.

Pulse findings and triage incidents are overlaid as their own nodes, connected to the resources they concern (AFFECTS, CAUSED_BY, RELATED_TO), so a resource's operational history is visible alongside its topology. Nodes and edges are versioned over time — the graph knows what changed and when, which powers drift views.

How mapping runs

The topology mapper is a scheduled discovery agent (daily by default). It scans your environment through your installed read-only cloud integrations, discovers compute, databases, networking, storage, and queues, and infers relationships from signals such as security groups, target groups, IAM roles, environment variables, and tags. Completed mapper runs appear in the Mission Control activity feed.

Using it

Graph explorer (web)

Open /graph. The explorer offers:

  • Display — switch between the Graph canvas and a Table view (see below); your choice is remembered per browser.
  • Search across your resources.
  • Modes — Topology, Blast (click a node to project its blast radius), Risks (single points of failure and toxic combinations), and Drift (what was added, removed, or changed over a selectable window).
  • Overlays — Health and Critical, plus toggles for stale and resolved items.
  • Grouping — by Cloud, Project, Region, Type, Layers, or free layout.
  • Stats bar — resources, dependencies, regions, health, stale count.
  • Node drawer — click a node (or a table row) for its properties, health (open findings), and blast radius.

Table view

The table shows exactly the resources the canvas would — the current drill-down scope and all active filters carry over — as one row per resource, with no visual aggregation. It's built for inventory work:

  • Nine sortable columns — Name, Type, Cloud, Project, Region, Health, Open Findings, Deps, Last Seen. Click a header to sort; the sort persists per browser.
  • Quick filter — type to filter across name, type, cloud, region, and project; click a Type cell to show only that type (a chip in the toolbar clears it).
  • Dependency drill-in — click a row's Deps count to expand its mapped edges inline: direction, relation, and neighbor; click a neighbor to jump to its row.
  • Export CSV — downloads the current filtered and sorted rows (not the whole graph) with resource, health, findings, dependency and last-seen columns; opens cleanly in Excel and Sheets.
  • Keyboard navigation — ↑/↓ to move, Enter to open the detail drawer, Escape to close it.
  • Rows for resources not seen in the last 14 days carry a stale tag — the same threshold as the canvas Stale toggle.

Search, Blast and Risks work in table mode too: search jumps to the row (widening any filter hiding it), and risk cards focus the affected resource's row.

In chat

The chat agent has graph tools; ask questions like:

  • "What services depend on the payments database?"
  • "What would break if prod-db goes down?" — blast radius, organized by dependency depth (default traversal depth 5).
  • "Find all single points of failure" / "any toxic combinations?"
  • "What are the most critical services?" — ranked by how many services depend on them.
  • "What changed in infrastructure since last week?"

The agent can also pull the full operational context around a resource — findings and incidents on it and its neighbors — before investigating.

Configuration

Setting Where Notes
Knowledge graph Team page → Infrastructure Knowledge Graph toggle Tenant admins only; on plans that include the graph
Plan — Requires the Growth plan or above
Mapper schedule Pulse agent schedule Daily by default
Blast radius depth Chat tool parameter Default 5 levels

Disabling the toggle turns off graph mapping and cross-domain linking for the tenant.

Troubleshooting

  • Graph is empty — the mapper has not completed a run yet, or no cloud integrations are installed. Install a read-only cloud tool and wait for the next scheduled run.
  • A resource is missing — the mapper only sees what your integrations can list; check the relevant tool is installed and active. DXDT also filters out structurally junk names at write time (aggregates like "and 3 more", placeholders, and extremely long space-heavy strings) to keep the graph clean — an unusually verbose resource name can be caught by this filter, so prefer concise resource naming.
  • Blast radius returns nothing — the resource ID or name must match a mapped node; use search in the explorer to find the exact name.
  • Stale nodes — resources not observed by mapper runs for 14 days are flagged stale, not deleted; use the Stale toggle to show or hide them. (The one exception: operators can explicitly expire structurally junk nodes as part of graph hygiene.)
  • Resources under an "Unknown" cloud or region — the integration reported no usable value for that facet; DXDT groups them under Unknown rather than inventing a bucket from placeholder values.