Strattum Knowledge Documents, decisions, runbooks, and validated answers converge into a live base — versioned, governed, and citable. Every question answered by the team is reabsorbed as an indexed asset, ready for the next decision.
The difference between a document repository and a knowledge layer lies in three properties: it grows on its own, maintains provenance, and responds with grounding. Strattum delivers all three.
Every answer validated by the team is reabsorbed as an indexed fragment. The base does not stagnate between projects — it deepens with every cycle of use.
Every answer carries a citation anchored to the document, version, and originating paragraph. Audit stops being a quarterly project and becomes a native property.
ACL, classification, and retention follow the client policy. Whoever cannot read the contract in SharePoint cannot access the excerpt via Strattum — no permission mirroring, no shadow IT.
The layer detects contradictions between documents, outdated policies, and coverage gaps before users encounter them. Base quality becomes an observable metric, not reactive auditing.
Ingestion that respects provenance, retrieval that respects governance, composition that respects usage history. The cycle closes on itself.
Official connectors read SharePoint, Confluence, Drive, Notion, repositories, and transactional databases. Each document enters with origin, version, author, and classification metadata preserved.
The layer combines vector, lexical, and entity graph search. Every query passes through the ACL filter before ranking — not after.
A team-validated answer becomes an indexed fragment, linked to the questions that originated it. The base grows through use, not through the next round of manual ingestion.
RAG answers today's question with yesterday's document. It works for prototypes; it stalls at enterprise scale. Strattum's layer treats the base as a live asset — with governance, provenance, and continuous composition from the source.
| Dimension | Classical RAG | Strattum Knowledge |
|---|---|---|
| Mental model | Retrieval pipeline | Versioned knowledge layer |
| Base growth | Manual, via re-ingestion | Continuous, via answer reabsorption |
| Provenance | Source document | Document, version, paragraph, and author |
| Permission | Post-filter on result | Pre-filter on index via inherited ACL |
| Base quality | Periodic audit | Continuous contradiction and gap linting |
| Validated answer | Disappears in chat history | Becomes an indexed, citable fragment |
| Multi-source | Chunk concatenation | Entity graph across sources |
| Operations | ML notebook | Platform with SLA and observability |
Andrej Karpathy described this category's destination as an "LLM Wiki" — a base combining human editing, automatic model contribution, and page-level governance. It is the reference architecture for corporate knowledge in the age of agents. Strattum Knowledge is that architecture, packaged with client ACL, client VPC deployment, and audit that survives a Risk Committee.
SharePoint, Confluence, Drive, Notion, Git, S3, transactional databases, and ERPs. Tenant OAuth, no intermediary credentials.
Dense, sparse, and entity graph retrieval, with domain-configurable reranking. No mandatory fine-tuning to enter production.
Permission inherited from source and applied before ranking. Permission revoked at source exits the index in the next cycle, without manual operation.
Every response returns with anchor to paragraph, version, and document. Click leads directly to the authoritative source, not an intermediate summary.
Revised document enters as a new version, does not replace the prior one. Old responses remain reproducible for audit and investigation.
Continuous detection of fragment contradictions, outdated policies, and coverage gaps. Base quality becomes a platform metric.
Telemetry by question, fragment, source, and user. Which documents support which decisions, with configurable retention by data class.
On-prem or BYOC in the client's cloud. Data does not transit to Strattum's cloud. Pluggable inference model — Bedrock, Azure OpenAI, Vertex, or self-hosted.
Analysts query policy, technical opinions, and internal precedents in a single layer. Every threshold decision carries a traceable citation to the document in the version current on the day of the operation.
Maintenance procedures, downtime history, and process adjustments remain available for the next shift and the next engineer. Senior operator tacit knowledge becomes an indexed, citable, versioned fragment.
Central and branch offices respond to members with the same product base, underwriting rules, and operational procedures. Divergence between branch and headquarters becomes a lint alert, not an ombudsman complaint.
Memory has entities and relationships. Knowledge has documents. Together, complete context.
Explore Memory Graph →
Skills use Knowledge to answer questions that depend on internal documentation.
Explore Skills →
Every access to sensitive documents is logged and respects original source permissions.
Explore Governance →The layer installs in your infrastructure and enters production in weeks, not quarters. Start with the source already mature in your environment — SharePoint, Confluence, Drive, or a repository — and expand at your operation's pace.
