Multi-agent DAG orchestration, purpose-built for enterprise engineering teams Learn more →

Knowledge & platform intelligence (Cortex)

Last updated 2026-05-25·8 min read

Cendriix Cortex is the unified knowledge and platform-intelligence layer behind the run loop. It builds a knowledge graph across your estate, Kubernetes, GitHub, observability, ticketing, and answers the question every SRE asks before a change: what breaks if this fails?

Cortex is pre-beta. CPT entity resolution and the blast-radius API are implemented in code; production deploy to prism-api.example.com is tracked as PRISM-P3. See the Cortex product page for the full overview and design partner program.

Overview

Modern estates describe the same logical service under different names on every platform. Cortex's patent-pending CPT Engine (Convergent Perspective Topology) resolves those aliases into one canonical graph node using structure and topology, not brittle string-matching, then exposes blast-radius queries to Cendriix agents and approval gates.

How it helps

Before an agent merges, deploys, or restarts a service, Cortex surfaces downstream impact, direct dependents, transitive rings, risk score, and shortest impact path. Operators see the same context in Infrastructure Intelligence (/infrastructure); agents call the graph via the cendriix.graph tool.

infrastructure intelligence · illustrativetopology
payments-apicheckout-svcledger-dbmobile-bfffraud-workerapi-gateway
risk score0.69

Same experience as /infrastructure in the operator console, not a public architecture diagram. Compliance and health panels remain illustrative until live ingest ships.

CPT engine

Five resolution passes merge cross-platform entities: exact name, normalized name, shared labels, fuzzy match, and topological fingerprint similarity (Hungarian assignment over a multi-dimensional fingerprint). The result is one canonical entity per real-world service.

  • Structural (35%), degree, neighbor types, clustering
  • Resource (25%), ports, protocols, resource class
  • Semantic (20%), 384-dim embedding (MiniLM)
  • Behavioral (10%), traffic and error buckets
  • Temporal (10%), age and change velocity

Blast radius

Given any entity in the graph, Cortex runs BFS traversal to produce impact rings, ranks entities by downstream fan-out, and traces shortest paths between source and target. The backend exposes this at GET /api/v1/blast-radius/:id; the Cendriix backend forwards queries with JWT passthrough for tenant isolation.

Platform integration

Cortex is not a separate product, it powers the run loop. Workflow composer runs an L3 blast-radius check before approval; the agent runtime queries impact from live runs. Graph answers in chat surface through the Cendra copilot in the Cendriix app (PRISM-P5), Cortex does not ship a separate in-product chatbot.

  • /infrastructure, topology graph + connector sync
  • /workflows/dag, optional blast-radius overlay per DAG node
  • CendraDock /graph <query>, PRISM-P5 graph Q&A

Availability

  • Live connectors: Kubernetes and GitHub (read-only ingest). Sixteen platforms in the catalogue; remaining connectors are design-partner priority.
  • Production URL: prism-api.example.com, planned, not deployed. Local Docker demo available for design partners (make dev + make seed).
  • Cendra copilot (Cendriix app): graph Q&A via PRISM-P5, no in-Cortex chatbot. Compliance and health UI panels use illustrative data until live ingest ships.