First Class Modules

Knowledge Core

Turns tenant knowledge into owned, revisioned, reviewable records that connect to business context, answer questions, support permission-aware AI/RAG, and learn safely from resolved support work.

Use Knowledge Core for policies, procedures, product help, FAQs, playbooks, support resolutions, and other maintained reference material whose ownership, revision, freshness, audience, and AI use must remain explicit.

What it does

Turns tenant knowledge into owned, revisioned, reviewable records that connect to business context, answer questions, support permission-aware AI/RAG, and learn safely from resolved support work.

Key capabilities

• Create Draft knowledge and publish or archive it through permission-checked lifecycle actions.
• Preserve immutable revisions, publication history, authority, AI eligibility, and optimistic-concurrency evidence.
• Assign one primary accountable owner to a user or DataRole and schedule recurring review work.
• Relate knowledge to any authorized RecordId and retrieve bounded contextual knowledge without copying the business record.
• Capture unanswered questions, repeated occurrences, usage, feedback, and secured knowledge-gap signals.
• Register only Published, AI-eligible revisions for secured semantic/multimodal retrieval with exact citations.
• Bring current support knowledge into fenced GoClaw Inbox context and turn resolved cases into sanitized Draft candidates.

Ownership, revisions, and lifecycle

Each item remains one canonical Record and Knowledge Core item. Its lifecycle is Draft, Published, or Archived; publication and archive are explicit permission-checked actions. Editing creates an immutable revision instead of erasing the evidence behind an earlier answer or citation. Stale browser updates are rejected through the item's expected update timestamp.

An active primary owner is either a named user or a DataRole, never an ambiguous mixture. Ownership makes review and escalation accountable without granting the owner access beyond normal tenant, DataRole, record, field, and location rules.

Reviews and freshness

A review policy sets the interval, due-soon window, stale threshold, and whether a review Task is created. Due work appears through the existing Tasks and Universal Inbox attention surfaces. An authorized reviewer can confirm the current revision, revise and republish it, reassign the review, or archive the item. Confirming current schedules the next review without pretending the content changed.

Freshness is evidence, not decoration. Current, due-soon, due, and stale states help users and AI distinguish maintained guidance from material that needs review. Stale content does not silently become authoritative merely because it still matches a search.

Relations and contextual knowledge

Knowledge can be linked to a customer, product, case, work order, conversation, file, or any other authorized RecordId. Related and contextual reads traverse the existing secured record graph within bounded depth and result limits. A relation improves context; it never grants access to either endpoint.

Questions, feedback, and knowledge gaps

When a user or GoClaw cannot find authoritative evidence, Knowledge Core records a question instead of fabricating an answer. Repeated occurrences are grouped inside the same tenant and may be assigned, answered, closed, or converted into a normal GeneratedCandidate Draft for review.

Usage and feedback such as helpful, outdated, or incorrect remain linked to the visible item and context. The health view combines repeated questions, weak/no-evidence retrieval, negative feedback, stale high-use sources, overdue reviews, and missing owners into secured gap signals. Aggregate counts never reveal an inaccessible question or source.

Secured AI, RAG, and citations

Only an explicitly Published and AI-eligible revision enters the canonical AI source, document-element, chunk, visual-index, and embedding pipeline. Tenant, CanAiReadRecords, DataRole, location, record, and field security are applied before model context is assembled and again when stored record identities are returned.

Answers retain the Knowledge item, immutable revision, authority, publication/review freshness, source, page, and visual evidence needed for a citation. Contextual record links can improve ranking, but relevance never grants access. Draft, archived, deleted, AI-excluded, or unauthorized material is absent from normal retrieval.

Universal Inbox, GoClaw, and support learning

Universal Inbox remains the support attention layer. GoClaw must hold the current claim/lease and exact ActivityVersion before assembling secured support context. Customer-visible citations include only support-eligible sources the represented user may disclose; internal-only knowledge is omitted. Read-only diagnostics can describe current account state, while RAG explains how the system should work. Neither authorizes a mutation.

After a case is Resolved or Closed, the controlled learning path accepts structured symptom, product area, likely cause, diagnostic steps, verified resolution, and affected-version fields—not the raw transcript. The support-resolution-v1 sanitizer removes credentials, secrets, PII, and tenant-specific identifiers. The result is an idempotent SupportResolution Draft with GeneratedCandidate authority, a source Inbox/version link, one owner, and a review Task. It remains unavailable to normal RAG until an authorized human reviews, marks it eligible, and publishes it.

OpsAtlas — AI Operational Knowledge Core

Your company’s operating knowledge, mapped and available on demand.

OpsAtlas is the thirteenth Build This product pattern and a complete product experience built on Knowledge Core. It turns SOPs, policies, manuals, training, system documentation, and hard-won tribal knowledge into maintained operating memory for employees and permission-bound AI assistants.

What OpsAtlas can do

• Ingest approved SOPs, policies, manuals, training material, system documentation, and tribal knowledge.
• Chunk, embed, version, and relate that knowledge while retaining source and revision evidence.
• Respect department, DataRole, location, record, field, and CanAiReadRecords permissions before retrieval or model context.
• Give employees evidence-backed answers to How do we do this? with exact sources, revisions, and freshness.
• Launch the appropriate BuildWithHQ workflow from an answer while preserving action permission, approval, and current-context checks.
• Give GoClaw agents and approved MCP tools the permitted operational context they need before they plan or act.
• Reduce repetitive internal questions and accelerate employee onboarding.
• Standardize procedures across teams and locations without flattening their access boundaries.
• Preserve institutional knowledge when employees, contractors, or specialists leave.
• Identify unanswered questions, weak answers, stale sources, negative feedback, and documentation gaps.

The continuous knowledge flywheel

Capture → Understand → Retrieve → Act + Improve is continuous: each real question or support outcome can reveal what should be reviewed, clarified, or added next without automatically promoting private or generated material.

OpsAtlas and CompanyIQ answer different questions

The two products can share BuildWithHQ’s secured records, relationships, AI retrieval, and action foundations. OpsAtlas organizes the business’s operating memory; CompanyIQ uses permitted business evidence to understand change and opportunity.

Common uses

• Owned product documentation, policies, procedures, FAQs, and employee help.
• Customer-safe help centers and support answers with exact current citations.
• Contextual playbooks linked to customers, products, cases, work orders, files, and conversations.
• Permission-aware AI/RAG sources whose authority and freshness must be inspectable.
• A controlled support learning loop that turns reusable resolutions into reviewable Drafts.

How it connects

Knowledge Core composes with Files and Web Links for source evidence, Record Relations for business context, Tasks and Universal Inbox for review/support attention, GoClaw for fenced assistance, and the existing AI ingestion/retrieval chain for cited answers. The Knowledge item and immutable revision remain the source of truth throughout.

Where applicable, its records use the universal RecordId conventions so they can participate in secured relationships, activity history, favorites, dynamic fields, notifications, Inbox attention, and global search without copying the source record.

Security and data boundary

Every list, item, relation, review, question, feedback, support-context, and AI path derives the SaaS app, tenant, user, DataRole, location, record, field, and permission scope from verified server identity. CanAiReadRecords is required before AI context. Publication, ownership, review, support learning, and archive actions have separate permission checks; customer-visible context excludes internal-only sources.

• The authenticated service derives the SaaS app, tenant account, user, DataRole, and location scope; browser identifiers are never authorization proof.
• The page editor composes React components with validated data bindings. Those bindings call typed runtime APIs, whose application services execute reviewed stored procedures.
• List, search, detail, relation, activity, favorite, and write operations reapply their required server-side permissions.

Add it to an app

1. Open the SaaS app in the Developer Console and add the validated module.knowledge-articles block to the pages and existing user-type menus that need Knowledge Core.
2. Create collections around reader/support journeys and assign their audience, default owner, review policy, and existing DataRole/location visibility.
3. Create or migrate source items as Drafts, confirm their authority and AI/support eligibility, then publish through the reviewed lifecycle.
4. Link knowledge to relevant Records, configure the review worker/Tasks and Inbox attention, and enable GoClaw support context only for permitted users and queues.
5. Test Draft/Published/Archived visibility, revision restore, stale updates, owner review, questions/feedback, cross-tenant/location denial, AI-read denial, internal-source exclusion, exact citations, stale Inbox activity, and Draft support learning before publishing an exact app version.