Skip to main content

// intelligence layer

WAIS

Westmarch Adaptive Intelligence System · pronounced “Ways”

Every CygNet system is configured differently. WAIS learns yours.

Corrections become candidate lessons. Lessons reinforce, decay, and graduate into site conventions. Senior administrators promote knowledge faster. A hybrid retrieval pipeline means the right knowledge surfaces at the right moment — and when you deploy WAIS centrally, every administrator's corrections benefit every other administrator on your site immediately.

the name

Named for the Red Book of Westmarch

The Red Book of Westmarch is the chronicle that compiled the hard-won knowledge of many journeys, written and added to over generations. WAIS is your CygNet's growing chronicle. The name “Ways” carries a double meaning: the ways of your CygNet system that WAIS learns, and the paths (ways) through knowledge that retrieval traverses.

// what WAIS encompasses

Six layers, one knowledge system

Each layer addresses a specific failure mode of static, rule-based knowledge in operational systems.

Multi-path retrieval
Structured ontology lookup runs first. Lexical and semantic recall layer on top. Results are fused and reranked for relevance to the actual question. Semantic search is additive — never the only truth path.
Adaptive learning loop
Failure-then-success patterns are detected automatically from real tool call sequences — no user flagging required. Lessons are captured with scope, provenance, and starting confidence, then evolve as the system observes more.
Knowledge graph
Entities, relations, and pattern generalization. Neighborhood-aware retrieval before entity operations. Entity-level corrections lift to class-level conventions over time.
Golden rules
Immutable product-level truths shipped with WAIS. Always retrieved for matching queries. Never decay. Your team can't accidentally argue with them.
Authority-weighted reinforcement
Admin / Operator / Observer roles weight correction confidence. Senior admins promote knowledge faster. Knowledge carries provenance, not just text.
Cross-instance site learning
Centralize WAIS and every administrator's corrections benefit every other administrator on your site immediately. This is the strategic moat.

// the flywheel

How WAIS gets smarter at your site

Every CygNet system is configured differently. WAIS observes how your team uses it, captures the corrections, and grows the chronicle — so the next administrator gets the answer instead of relearning it.

  1. 01

    Use

    Engineer runs a skill or asks Narya to handle a task.

  2. 02

    Correction

    Failure-then-success patterns are detected automatically — without the user having to flag them.

  3. 03

    Lesson

    A candidate lesson is captured and tagged with scope, provenance, and a starting confidence.

  4. 04

    Reinforcement

    Repeated observations grow confidence. Senior administrators carry more weight, so trusted knowledge promotes faster.

  5. 05

    Convention

    Reinforced lessons graduate into site conventions. Entity-level corrections lift to class-level rules over time.

  6. 06

    Skill

    Promoted into a reusable Skill so every administrator gets the answer next time — not just the one who learned it.

// the moat

Centralize WAIS, and every admin's knowledge becomes every admin's knowledge.

Run one Westmarch server for the team and every correction, every convention, every learned alias is immediately available to every other administrator on your site. Not next quarter. Not after a model retrain. Now.

// deployment

Three modes, one connection string

Same WAIS API, same schema, same code path. The only difference between modes is where the knowledge store lives and which cloud calls are enabled.

Local — single machine

Eval and single-admin sites

Narya installer can provision the WAIS knowledge store on the workstation. The intelligence layer runs as a local service. Everything stays on one machine. Good for evaluation.

Centralized — the moat

Production, multi-admin sites

Run one Westmarch server — your existing managed database service, or one we help you stand up. Every Narya instance points to it. All learned knowledge is shared instantly.

Strict-OT — air-gapped

When cloud AI is off the table

Same schema, same API. Cloud-only retrieval steps are disabled cleanly; structured and lexical retrieval still work. Storage and architecture are identical regardless of OT policy.

what stays in your environment

Customer ontology and memory remain local and authoritative

Site names, device IDs, facility tags, point tags, raw notes, raw tool outputs, and changeset payloads stay on customer infrastructure by default. Any external retrieval calls are minimal, scoped, and can be disabled wholesale in strict-OT mode.

Future Global Anonymized Generative Learning — cross-customer product intelligence — is contractually opt-in only, anonymized at the source, and can never include raw customer identifiers by default.

how knowledge is stored

Typed records · raw evidence · semantic index

Durable typed knowledge — entities, relations, lessons, conventions, workflow runs, cases, approvals — lives in queryable records. Raw evidence and flexible payloads sit alongside, fully retrievable. Semantic indexing attaches to the memories that benefit from it. We don't index every chat turn or every changeset blob — accuracy and consistency over cost at every decision point.

// ready?

Put WAIS in front of your team

WAIS ships inside Narya Command. Get the beta now and let your site's chronicle start growing.