The vocabulary in this space is crowded and shifting. These are the terms we use, and what we mean by them.
Agent-ready. Data an AI agent can use correctly without a person in the loop to interpret it: curated to what matters, explained, scoped to the caller’s permissions, and delivered with guidance on use. Distinct from AI-ready, which usually means clean and well-structured; data can be perfectly clean and still carry no semantics.
Semantic layer. A layer between a data store and whatever is asking it questions, holding the business definitions so consumers don’t reconstruct them. The category is thirty years old in its BI form. Most semantic layers — Snowflake semantic views, Databricks metric views, dbt, Cube, Tableau Semantics — are metric-level: they define measures and dimensions over a warehouse for aggregate queries. ContextWorks is a record-level semantic layer for Salesforce CRM data: it defines what an agent receives when it opens one record. Complementary, not competing.
Context layer. What a semantic layer produces when the consumer is an agent: the assembled, explained, governed context delivered at request time. We use it for the output. We don’t lean on it as a differentiator, because half the data industry adopted the phrase in 2026.
Entity. A curated view of one Salesforce object: fields in order, rollups, filtered related lists, parent, enrichments, instructions. Three depth levels from one definition.
Element. One item inside an entity: a field, a formula, a rollup, a related list, an enrichment.
Enrichment. A computed element from your own code — an Apex provider, an invocable action, a Flow, a declared Data 360 element — composed into an entity by configuration.
Brief. Institutional knowledge — segmentation, what a stage means, brand voice, policy — as a governed, versioned artifact, injected per domain.
Skill. A procedure an agent follows for recurring work, referencing entities and briefs by name. An artifact any agent can run, not part of one agent.
Domain. A purpose-built access point: a set of entities, briefs, and skills composed for one job, owned by a business function. Every retrieval names a domain.
Pragmatics. Borrowed from linguistics: meaning in use, as opposed to semantics, meaning in isolation. A description says what a field means. A pragmatic instruction says how to use the value — “below 60 warrants outreach; internal only.” Attached per field, per entity, per domain. The thing we think nobody else ships.
Enforcement, not filtering. Permissions applied to the query that assembles the context, as the calling user, rather than to the result afterward. No code path where a value is fetched and then hidden.
Apache Ossie. Open Semantic Interchange, now incubating at Apache: a vendor-neutral format for semantic definitions, backed by Snowflake, Salesforce, Databricks, dbt and others. ContextWorks exports to it, so definitions can leave the system as data. It’s dataset- and metric-centric today; record-level constructs are a gap we’d like to help close.
🤠 Made in Montana. World-class software from a magical place. Building on Salesforce since 2017.
