What is a tenant knowledge graph?

Why a graph rather than a transcript

Salesforce work is relational. A field belongs to an object, is written by a flow, is read by an Apex class, is exposed by a permission set, and was introduced by a decision recorded somewhere else. Answering a practical question - why is this flow failing for one region - means traversing those relationships, not searching text.

A conversation transcript stores what was said, in order. It cannot answer which permission sets grant access to a field, because that relationship was never modelled. A graph stores the entities and the edges between them, so the connections are queryable.

What RplAI stores, in four layers

RplAI organises tenant knowledge into four layers, each answering a different kind of question:

• L1 - Org metadata: objects, fields, flows and permissions, read from the real environments rather than from documentation.
• L2 - Repository awareness: the codebase, branches and the conventions the team actually follows.
• L3 - Tenant knowledge graph: decisions, documentation and drift, connected across the whole tenant.
• L4 - Agent memory: session context and feedback signals from previous executions.

Tenant scoping

The word tenant is doing real work in the term. The graph is bounded to one customer environment, which is what makes it safe to store specifics - field names, business rules, internal decisions - that must never leak into another customer’s answers.

For a consulting organisation working across many client orgs, that boundary is the difference between an AI system that can be used on client work and one that cannot.

Why persistence changes the output

When knowledge persists, each execution can add to it. A resolved incident records its root cause; a generated flow records the convention it followed; a rejected suggestion records why it was rejected.

The practical effect is that later requests start further along than earlier ones. Without persistence, request ten is identical in cost to request one, because nothing learned in between was retained.