Natural language to agent intent. Grounded in your graph.
Cendra is the conversation interface to the Cendriix platform. It maps your plain-language instructions to structured agent intents, querying the Cortex knowledge graph to ground every instruction in your actual systems — not a generic LLM context window.
Instructions that become executable plans.
Cendra does not generate free-text responses. It resolves your instruction against the Cortex knowledge graph and outputs a typed execution plan — each step bound to real entities in your systems.
Intent resolution
Cendra parses your instruction and resolves it against Cortex entities. "Fix the PagerDuty alert" becomes a structured plan with blast-radius context attached — not a vague LLM response.
Graph-grounded responses
Every Cendra answer pulls from the Cortex knowledge graph — your actual Jira tickets, Kubernetes state, and Datadog metrics. No hallucination about systems Cendra has not ingested.
Multi-turn context
Cendra maintains conversation context across turns. "Now update that service's runbook too" builds on the previous turn's Cortex entity resolution without re-querying.
A2A-ready routing
Cendra identifies when sub-tasks need specialist agents: a code-writing agent, a test runner, a deploy agent. It routes and hands off context automatically.
Structured plan output
Cendra's intent resolution produces a structured execution plan, not a freeform text response. Every step is a typed agent action with Cortex entity bindings — executable, not aspirational.
Disambiguation with evidence
When an instruction is ambiguous ("deploy the service"), Cendra surfaces the candidates from the knowledge graph and asks you to confirm — with blast-radius scores attached so you can choose with context.
Not a chatbot. An intent engine.
Grounded, not hallucinated
Cendra only surfaces entities that exist in the Cortex knowledge graph. It cannot invent a service, ticket, or deployment that it has not ingested.
Structured output, not prose
Every Cendra response is a typed plan — a sequence of agent steps with entity bindings. Agents execute the plan; they do not parse a freeform paragraph.
Conversation state in the graph
Cendra's conversation context is stored as a Cortex entity, not an ephemeral in-memory session. Every turn is auditable and reproducible.
Role-scoped access
Cendra's answers are filtered by the authenticated user's RBAC policy. It will not surface entities the user's role is not permitted to see.
See intent resolution in action.
Graph-grounded instructions. Structured execution plans. No hallucination about your systems.