Knowledge grounding
How Campaign Builder keeps campaign context usable by agents.
Campaign Builder's Knowledge v2 layer stores reusable brand, audience, market, compliance, campaign, data, and agent context with taxonomy validation, source metadata, embeddings, visibility scope, and governance fields before teams move into SFMC execution.
Grounding workflow
From reusable context to better campaign work.
The Knowledge routes are not another production channel. They make account and workspace context searchable, classified, and reviewable so agent outputs can reflect the campaign team's real operating rules.
Define the workspace taxonomy
Knowledge v2 exposes effective, workspace, validate, save, and reset taxonomy routes. The default taxonomy covers business, audience, content, campaign, data, compliance, and agent categories with allowed knowledge and source types.
Ingest reusable context
Authenticated users can add manual entries or ingest public HTML URLs. URL ingestion extracts readable article, main, body, heading, paragraph, list, and quote text, chunks it, stores source metadata, and rejects non-public URLs.
Search with readable scope
Knowledge search embeds the query with text-embedding-3-small, runs vector search where available, falls back to text search when vector search is unavailable, and always applies account, workspace, or private readable filters.
Feed grounded workflows
Campaign Agent, Content Agent, Advisor, and Journey flows can use readable workspace knowledge as grounding, while Bruno and Factory still require their own entitlements and confirmed SFMC execution boundaries.
Implementation map
What evaluators can trace to Knowledge routes.
Knowledge is mounted as an entitlement-gated tool route under `/api/v1/mcp/knowledge`. The implementation validates entries against the active taxonomy and applies readable scope before records are returned.
`GET /api/v1/mcp/knowledge/taxonomy/effective` returns the active workspace taxonomy or the generic default taxonomy for the authenticated account and workspace.
`PUT /api/v1/mcp/knowledge/taxonomy/profile`, `POST /taxonomy/validate`, and `POST /taxonomy/reset-default` let teams validate, save, or reset workspace taxonomy profiles.
`POST /api/v1/mcp/knowledge` validates classification, source type, governance fields, applicability fields, visibility, embeddings, and required taxonomy fields before saving an entry.
`POST /api/v1/mcp/knowledge/ingest-url` fetches public HTML, extracts readable chunks, embeds each chunk, deletes prior entries for the same docId, and stores source title, URL, page, and section metadata.
`POST /api/v1/mcp/knowledge/search` supports classification filters, governance status, docId, modality, vector search, and text-search fallback under readable account and workspace scope.
`GET /api/v1/mcp/knowledge/recent`, `GET /:id`, `PUT /:id`, and `DELETE /:id` only return or mutate knowledge records readable by the authenticated actor.
Evaluator evidence
A good pilot should show how context is scoped and governed.
Ask for evidence that the same source, classification, visibility, and governance fields used in storage are visible during review. That prevents generic AI copy from being mistaken for approved account knowledge.
Scope evidence
Readable filters include account knowledge, workspace knowledge for the active workspace, and private knowledge for the current owner user. Account knowledge uses a null workspaceId; workspace and private knowledge keep workspace scope.
Taxonomy evidence
Validation checks enabled categories and types, allowed knowledge types, allowed source types, and required fields for the selected taxonomy type before entries are accepted.
Source evidence
Each entry keeps source type, title, URL, import time, sourceId, legacyId, chunk docId, modality, page, section, tags, and applicability metadata for later review.
Governance evidence
Knowledge entries include governance status, risk level, valid-from and valid-until dates, requiresHumanReview, and verified flags so teams can separate draft guidance from approved reusable context.
Technical evaluator FAQ
Where knowledge helps, and where SFMC review still matters.
Treat Knowledge as the grounding layer for better campaign decisions. It should improve the inputs to execution, not hide or replace Marketing Cloud review.
Is Knowledge grounding the same as SFMC execution?
No. Knowledge grounding supplies account and workspace context for planning, content, recommendations, and journey preparation. SFMC writes still run through confirmed specialist, Bruno, or Factory execution paths.
Can every workspace read every knowledge entry?
No. Repository filters restrict reads to the authenticated account plus account, active workspace, or current-user private visibility. Workspace knowledge requires the matching workspaceId.
How should evaluators test source quality?
Review taxonomy validation, source metadata, chunk docId, page and section fields, governance status, and whether search results cite readable account or workspace records rather than unrelated documents.
Where does this improve SFMC campaign production?
Grounded knowledge helps Campaign Agent briefs, Advisor recommendations, Content Agent drafts, and Journey planning reflect approved brand, audience, market, compliance, and workflow context before reviewed SFMC artifacts are prepared.
Enterprise guardrails
Grounding stays scoped before SFMC work begins.
Knowledge grounding is useful because it is constrained: authenticated actor context, workspace-readable filters, taxonomy validation, source policy, and clear separation from confirmed SFMC writes.
Authenticated actor required
Knowledge routes require authenticated account and workspace context before taxonomy, recent entries, search, ingestion, create, update, or delete operations run.
Public URL policy
URL ingestion accepts only public http or https URLs and rejects private or non-HTML targets before content is chunked and embedded.
Grounding before execution
Knowledge improves the quality of plans and recommendations, but it does not approve sends, activate journeys, or bypass SFMC write confirmation.
Evaluation takeaway
Better grounding makes execution safer to review.
A buyer evaluating Campaign Builder should expect Knowledge to make campaign briefs, recommendations, content drafts, and journey planning more specific to the account. They should not expect Knowledge to send campaigns, activate journeys, or grant unrestricted SFMC write access.