Context Graph
Shared semantic memory, built for speed
A self-organizing graph of entities, relationships, events, and facts woven from your connected data. It lives in our native graph engine, built for agent memory: each query draws 2,668 tokens instead of 202,285.
How it works
Graph traversal, not flat retrieval
Starting from entities relevant to the query, the platform follows the graph's real relationships, including causal links, to assemble connected context.
Token economics
98.7% fewer prompt tokens per query
Graph-aware retrieval sends only the connected neighborhood relevant to the current query. The prompt carries the few entities and relationships that matter, not the entire knowledge base. At $3 per million input tokens, that is roughly $599 saved per 1,000 queries.
Raw corpus vs connected neighborhood
Graph-aware retrieval economics
Conventional RAG
0 tokens
Graph-aware retrieval
Actual proportional bar is tiny; callout marks compressed context packet.
0 tokens
0%fewer prompt tokens
Retrieves the relevant neighborhood of the graph.
Cost impact at scale
Tokens sent to the model per query
Conventional RAG
0 tokens
Wexa
0 tokens
98.7%
reduction per query
Pipeline
Five stages from query to context
Query arrives
Agent sends a natural-language query through the governed API.
Entity resolution
Platform identifies the entities relevant to the query from the context graph.
Graph traversal
Follows typed relationships (depends-on, derived-from, mentioned-in) to gather connected context.
Provenance stamping
Every entity that reaches an agent is stamped with source, timestamp, and confidence.
Minimal context delivered
Only the connected neighborhood reaches the prompt. Cost stays flat as knowledge grows.
Ontology
Self-organizing, human-reviewed
As connectors stream in at up to 5,588 writes a second, the platform discovers entities and relationships and scores its own confidence. Low-confidence or conflicting mappings route to a human review queue, so the semantic layer is both automatic and trustworthy: two departments' agents mean the same thing by "customer". Ontology Studio, a canvas editor, lets your team shape entity types and relationships directly.
Governed Tables
Structured records live beside the graph, with the same policy checks and audit trail on every read and write.
Structured data gets the same governance as the graph. Nothing lives outside policy.
Context policies
Per-agent scopes define which entities, relationships, and Tables each agent can draw into its prompt.
Blast radius is defined per agent before it ever runs.
See the context graph in action.
Book a 30-minute scoping call or send us a message. We respond within one business day.