DNA · SUBSTRATE

Context that compounds.

The typed-graph substrate Eyes and Cortex both sit on. Dependency-aware memory. Permission-aware retrieval. Every event becomes a node; every node carries its trace.

See Cortex →
+11%
judge-score lift over raw agents
internal benchmark, May 2026
0
edges without provenance
every edge cites its trace
1
graph for dev, eval, and prod
no translation layer
01

Substrate

Events, chunks, claims, objects, edges, semantic links, surface items — a typed graph that everything else (Eyes, Cortex, your judges, your guardrails) reads from.

02

Dependency-aware

DNA reads the dependency graph between artifacts, runs, and operators. Cortex surfaces the right context at the right hop — not flat embeddings.

03

Quality wedge

+11% judge-score lift over raw agents in internal benchmarks. The advantage compounds: context that started in dev shows up automatically when the same shape appears in prod.

Dependency-aware retrieval, across handoffs.

Most RAG ships flat embeddings. DNA reads the dependency graph between artifacts, runs, and operators — and surfaces the right context at the right hop. When the orchestrator delegates to a sub-agent, the relevant policy travels with the intent.

How it composes

Eyes writes to DNA. Cortex reads from DNA. Judges and guardrails ride along.

One substrate, two surfaces. Every event Eyes captures becomes a typed node DNA can reason over. Every Cortex answer cites the same graph. Findings travel back into guardrails — and the score keeps climbing.

See DNA on your own runs.

15 minutes. One workflow. No deck.

See Eyes →