Vector search
Finds passages that look like the question.
Good for finding the right document. But a similar passage is not an answer: it does not tell you whose it is, which rule it touches, what has changed since.Platform · AI Infrastructure
Your business knowledge, structured for an AI to query.
A knowledge graph connects your company’s data, processes, and tacit knowledge in a single structure: the most effective way for an AI to navigate data, which dramatically increases the speed and precision of answers and is easy to audit by design. Controlled access via MCP Server for all your agents and employees, and real-time intelligence via Events.
Turning information into knowledge
Fernando · advisor“And the kids, how are they doing?”
Ricardo · client“Rafael is leaving home, going to study abroad. At Berkeley.”
Fernando · advisor“You’ll miss him, I imagine. But it’s an investment for life, right?”
Ricardo · client“Absolutely. 5 years living abroad. He’ll learn a lot.”
01The concept
Why a graph, and not another document search.
Knowledge graph
Traverses entities and relationships: “which clients in this portfolio were affected by that rule change?”
A question no single passage answers on its own. It requires walking through client, portfolio, rule, and date, in the right order.RAG finds the source. The graph organizes the path. The two coexist: the graph provides the spine, search provides the recall.
02Knowledge Graph
Your business as a living model.
Entities, relationships, rules, and time: the entire operation represented in a way an AI can traverse and verify.
How it is born
From two raw materials: the unified data from the foundation and the business context (rules, exceptions, vocabulary, who decides what) captured in the Assessment with the people who live the operation.
How it is designed
The ontology, the grammar of this representation, is modeled with the people who know the operation from the inside. Without it, a large graph only organizes noise.
How it is verified
Every node keeps the path back to its source: a document, a system record, or a meeting excerpt with a timestamp. No answer leaves without an origin.
Example
“The expansion moved to August”, said in a meeting, becomes a node linked to project, deadline, and owner. When someone asks about the project in October, the sentence resurfaces, with the source right beside it.
03MCP Server
Context exposed with controlled access.
Agents and assistants query the graph through MCP, the Model Context Protocol, an open protocol. What each agent can see is your decision, on record.
Permissions inherited at query time
Role-based permissions hold inside the AI: advisor A’s agent cannot see advisor B’s portfolio. Not by mistake, not by creative prompting.
Each agent, its own boundary
Read access, tools, areas of the graph: each agent receives only what its role requires. Scope is a contract, not a convention.
Every query on record
Who asked, what was queried, when, with which answer. Compliance audits the AI the way it audits people.
Credentials out of the prompt
Keys and passwords never pass through the model. What the model does not see, the model cannot leak.
@assistant: how is client Almeida’s portfolio doing?Answer scoped to whoever asked · query recorded in the audit trailThe model is pluggable.
Switching models is a procurement decision, not a one-year project: the graph, the permissions, and the audit trail stay where they are. What changes is the engine, not the foundation.
04Event Bus
Intelligence the moment it happens.
From the data that just arrived to an alert for whoever needs to act, without waiting for Monday’s report.
The event comes in
Every connected source publishes to the bus: a transaction, a new record in the CRM, a meeting processed by VOX.
The condition evaluates against the graph
Rules defined with the business read the event with context: whose it is, which rule it touches, which deadline it affects.
The action fires
A custom alert in the team’s channel, a dashboard updated in real time via streaming (SSE), or a task for an agent.
Example · Compliance
An unusual transaction in a conservative-profile account. The owner gets the alert immediately, with the history and the rule right beside it, not weeks later in a report.
05Side by side
The three pieces, in one table.
| Component | What it does | Technical term | What changes in the operation |
|---|---|---|---|
| Knowledge Graph | Structures entities, relationships, rules, and time | Ontology · lineage | The AI answers with real context and cites the source |
| MCP Server | Exposes the graph to agents and assistants, with control | RBAC · scopes · audit trail | Each one sees only what it can; everything is recorded |
| Event Bus | Evaluates events against the graph and triggers actions | Events · SSE | The exception shows up immediately, not in the audit |
Compliance alerts
The exception shows up immediately, not in the audit.
See the case ↗Advisor IntelligenceThe day starts knowing which client to serve first. And why.
See the case ↗VOXThe meeting ends. What was agreed does not get lost.
See the case ↗Where this layer fits
Explore the AI Applications↗It consumes the Data Infrastructure base and serves the Applications. Contracted as SaaS.
06Next step
Want to see the graph with a sample of your data?
A technical conversation, with your scenario on the table. We bring the architecture; you bring the hard questions.
07Contact
Book 30 minutes
with the people who build.
A founder joins the call. You bring a real problem from your operation and leave with an honest read on where AI creates value. If we are not the right fit, we say that too.
The calendar opens right away. You pick the time and get the invite by email.
If you would rather write before booking, the email is contato@entitylabs.com.br.
What fits in 30 minutes
- You tell us which decision is stuck in your operation today.
- We show the layer applied to a case like yours.
- We tell you what can be done with the data you already have. And what cannot.
- If it makes sense, you leave with a written scope — timeline and success criteria included.