What ContextWorks adds to Salesforce.
The platform solved getting your agents to the data. What that data means is still yours to define.
Start with what's already there.
Context and semantic are doing a lot of work in the market right now, and several Salesforce features wear one of those words. These are the pieces the platform has, and what each one actually provides.
Hosted MCP Servers
Secure, governed, per-user access to CRM data for external AI tools. Every call runs as the authenticated user, through field-level security and sharing. This is the access problem, solved properly.
Data 360
The data platform. It unifies and models data from across your systems, and powers unstructured grounding, indexing and agent observability. Grounding an agent in CRM records does not require it.
Tableau Semantics
A real semantic layer, and a good one. It defines metrics once — what revenue or churn means — consistently for analytics and for agents. It is bound to Data 360 and oriented to metrics and aggregates.
Agentforce Observability
Session tracing over agent reasoning, so you can see how an agent arrived at an answer. Strong on the reasoning; it is not a view of the context payload assembled for a given record and user.
Field Descriptions
Salesforce's own MCP returns admin-authored field guidance to agents. That is real, and it validates the idea that a field needs an instruction attached to it.
Context Service
Worth naming because the word collides. This is an Industries developer feature — an in-memory data context that saves repeated database trips for Flow and the Business Rules Engine. It is a performance mechanism, unrelated to AI.
We use the platform rather than rebuild it.
ContextWorks is a Salesforce-native application. Permissions, sharing and the trust layer are not features we reimplemented — they are the platform's, and we run inside them.
| Field-level security and sharing | Every query runs in user mode. Enforcement is applied to the query, not as a filter afterward. |
| Hosted MCP Servers | Custom servers scoped to a ContextWorks domain, using the platform's own OAuth and registry. |
| Einstein Trust Layer | Agentforce delivery flows through it unchanged. |
| Prompt Builder, Flow and Apex | Grounding for templates, invocable actions for automation, and a public API for your own code. |
| Field metadata and page layouts | Entities initialize from your existing setup rather than a blank canvas. |
| Data 360 | Calculated Insights and DMOs surfaced as enrichments on a record. |
Full context, built by the people who know the data.
Not a copy of your CRM and not a model of your metrics — the layer in between, where a field gets a meaning, a definition gets an owner, and every AI tool reads the same thing.
Entities
The records, curated
A chosen set of fields per entity at three depths, each with an instruction attached — assembled into one payload with rollups over child records, filtered related lists and computed signals.
Briefs
The knowledge that isn't in any field
How the business segments, what a stage actually implies, brand voice, methodology, policy. Authored once by the function that owns it and injected wherever it applies.
Skills
How the work gets done
Procedure as a standalone artifact any agent can use. A skill references a named definition rather than describing a filter in prose, so the meaning can change underneath it and the skill keeps working.
Ownership
Maintained by the business
Each team maintains its own domain, with delegated administration, draft and active versions, and one-action rollback. No admin, no sandbox, no deployment.
One definition, every surface
Built once, read everywhere
MCP, Agentforce, Prompt Builder, Flow and Apex all read the same definitions — so the answer does not depend on which tool someone happened to open.
