Compile fragmented data into governed, actionable context.
Nexum compiles structured records and documents into a continuously maintained, typed, evidence-backed graph that people, applications, and AI agents can query, traverse, and trust.
Whether you are building Node.js applications, architecting enterprise systems, or designing Retrieval-Augmented Generation pipelines, Nexum turns disconnected organizational information into reusable knowledge.
The pure graph compiler
Nexum is built from the ground up as a platform-neutral TypeScript compiler, with the type safety and strict validation large-scale enterprise codebases need. It treats organizational knowledge as a compilation target, providing a normalized contract for your records and documents.
Embeddable SDK
Import @x12i/nexum directly into your Node.js or Fastify processes for offline processing and the lowest-latency execution — entirely independent of a database.
Deterministic extraction
Extract deterministic, typed graphs from JSON Schema, OpenAPI specifications, endpoint definitions, typed records, and Markdown structure.
Idempotent apply plans
The compiler never writes to a database. It evaluates a source snapshot against your registry and emits an evidence-backed apply plan — replace-contribution and remove-contribution actions — for your host system to execute.
Flexible deployment
Run Nexum in-process via the SDK, scale it as a stateless service with @x12i/nexum-service, or integrate directly into your persistence layer with @x12i/memorix-nexum-adapter.
Governed organizational context
Organizations do not lack data; they suffer from the risk and cost of repeatedly reconstructing context from disconnected systems. Nexum centralizes relationship meaning while preserving the authority of your source systems.
Conservative identity resolution
Strict, registered identity rules protect enterprise integrity. Ambiguous matches, shared names, or fuzzy AI inferences never trigger silent automated merges — they become durable review candidates for human stewardship.
Evidence and provenance
Every relationship is auditable. Nexum stores the source revision and exact evidence locator for every claim, giving a defensible lineage from operational reality to graph output.
Strict ACL safety
Sensitive facts are protected at the graph level. Authorization filtering applies to claims, edges, paths, counts, and explanations before computation is exposed — hidden relationships cannot leak through path existence or aggregated totals.
Incremental updates
Source updates and deletions replace only their specific contribution, so a shared entity or edge survives as long as another valid source still supports it.
Bounded, evidence-backed truth
Standard agents repeatedly retrieve raw chunks, notice missing context, trigger costly retrieval loops, and still cannot reliably explain their conclusions. Nexum provides a structured, verifiable knowledge layer for Model Context Protocol servers and RAG architectures.
Query compiled meaning
Instead of retrieving raw text, agents query an already-compiled graph for exact, typed relationships — drastically lowering token use, latency, and retrieval complexity.
Bounded subgraphs
Supply agents with small, narrowly scoped graph slices and paths, tailored to their specific task and authorization level.
The "why" path
Agents can inspect the why payload for high-impact claims to retrieve explicit evidence, so actions stay grounded in verifiable reality.
Evidence-constrained AI
When Nexum's optional AI extraction is enabled for prose, it runs under token budgets, required evidence spans, and provider tracking — invented quotes are strictly dropped.
High-value compilation use cases
Nexum turns raw inputs into explicitly connected, queryable domains.
| Domain | Inputs | Resulting context | Business value |
|---|---|---|---|
| Engineering architecture | Service records, Markdown, OpenAPI, endpoint definitions. | service → exposes → endpoint → uses_schema → schema | Safer releases and focused regression testing by identifying impacted consumers. |
| Operational incidents | Service assets, ownership records, runbooks, alerts. | incident → affects → service → owned_by → team | Faster incident routing and reduced context gathering during triage. |
| Security & compliance | Assets, findings, controls, policies, exceptions. | finding → affects → asset · control → protects → asset | Audit-ready traceability and clear risk ownership with evidence mapping. |