PT EN
Back to site

Create a context, a workspace and a panel — through the screens or the Captain

A new corpus on the platform (contracts, orders, reports, a statute) is born in three steps, none of which needs a developer: the data context (where the documents live), the workspace (what the team sees and which contexts it uses) and the panel (the list of that context's records). This guide shows the two paths that do exactly the same thing: clicking through the screens, or asking the DATTA Captain in plain language. The outcome is the same — the Captain only calls the same operations the screens call, and never runs without your confirmation.

The example used here is the Contratos context: business contract documents, each belonging to a Contrato record identified by contratoId, screened with the rules of the Civil Procedure Code.

Before you start: the data anchor

Every context whose documents belong to a record of their own must declare two fields, and this is where most problems are born:

  • Anchor entity — the record's name in the graph, capitalised and singular: Processo, Contrato, Pedido, Laudo.
  • Identifying property — that record's unique key: processoNumero, contratoId, numeroPedido.

That pair is what links every document chunk to the right record. Without it the document is indexed, shows up in search and chat, but screening finds no text ("no text to evaluate") and the panel lists nothing — with no error on screen. The Judicial Process, Registration Dossier, Credit Analysis and AML/CFT Investigation types already carry a default anchor; the Custom type — the one you use for any new corpus — asks you to declare it (both fields together, or neither).

The context's entity-extraction prompt must emit that entity with that property — one Contrato with contratoId per document, in the example. The platform seeds default prompts when the context is created; adjust them in SistemaContextos, Prompts e Regras section, if the corpus has its own entity. Type the property carefully: a typo (conrtatoNumero instead of contratoId) has the same effect as declaring nothing.

Path 1 — through the screens

Step 1: the context

  1. Open SistemaContextos and click + Novo Contexto.
  2. Display label: the name shown on the platform (Contratos). The internal name is derived from it and is what integrations use — leave it unless you have a strong reason; it does not change later without moving data.
  3. Context type: pick the category. For a new corpus, Custom plus the subtype (Contratos). For court cases, Judicial Process.
  4. Data identification: for Custom, fill Anchor entity (Contrato) and Identifying property (contratoId) — see the previous section. For types that already carry the anchor, check the pre-filled default.
  5. Description and, if any, Source URL (for a corpus fetched from a site). The description is stored in the context and comes back filled in when you open Editar. In a Judicial Process context, also fill in the procedural code database (cpc, clt, cpp…) — the database name, not the code's: left blank, "Gerar recurso" uses the Code of Civil Procedure; with a database that does not exist in Neo4j, it comes out without articles. In both cases the screen warns on save.
  6. Data sources: leave nothing selected so the platform creates the context's graph database and search index — the yellow notice shows the names that will be created. Select existing sources only when the corpus already lives in a registered database.
  7. Rule contexts applied in screening: tick the rule corpora screening must evaluate over these documents (Código de Processo Civil). With none, screening evaluates no rule at all — and the screen says so.
  8. Criar Contexto. The confirmation shows the database and index created; the Schema do Grafo, Embeddings, Bancos de Dados and Prompts e Regras sections of the context now exist on the same screen.

Once created, in the Context Visibility Matrix (top of the same screen) tick the features the context appears in — Upload PDF is what makes the context show up on the Upload screen. A new context is born ticked in the general features and unticked for upload when the installation restricts upload to some contexts; tick the cell and you are done, there is no save button.

To verify: send a document through Upload, choosing the context, then open SistemaContextos, Schema do Grafo: the anchor entity must be among the nodes. If it is not, the extraction prompt is not emitting it.

Step 2: the workspace

  1. Open SistemaWorkspaces and click Novo workspace. Enter the name (Contratos) and a description. The workspace is born empty.
  2. Funcionalidades tab: tick what the team will use — usually Upload de Documentos, Explorar Dados, Regras de Triagem and Triagem em Lote — and click Salvar funcionalidades. With the workspace active, the side menu shows those work features — and, always, Chat, System, Panels, Features, Documentation, DATTA Captain, the profile and the workspace switcher.
  3. Contextos de Dados tab: tick the context you created (Contratos). The screen shows which rule corpora it is connected to. Salvar contextos. Saving the contexts links no panel: a context with an anchor offers the anchor's triage panel (Painel de Contratos), and the screen reminds you to link it — in the Painéis tab, Vincular existentes, Nativo group. Link it if the team will use the triage; the declarative panel of step 3 is separate.
  4. Membros tab: who has access, with a role. A workspace without members — or with only platform administrators — is visible to everyone with read permission; the first member who is not an administrator restricts it, and the screen warns beforehand.

Step 3: the panel

  1. On the workspace's Painéis tab, click Novo painel, give it a title (Contratos) and Criar e abrir — the builder opens with the empty panel.
  2. At the top of the builder, the context selector already comes set to the workspace's context when it has a single one; with several, pick the right one. That selector decides which database the origin list's entities come from.
  3. Add elements from the left palette and configure each on the right:
    • Metric card: source Neo4j (grafo), origin = the anchor entity (Contrato), aggregation COUNT — the record total.
    • Filter: origin Contrato, field tipoContrato (or another entity field) — becomes a value list that filters the table.
    • Search: origin Contrato, searched fields contratoId, titulo.
    • Table: origin Contrato, columns with the fields that matter (contratoId, titulo, tipoContrato, dataAssinatura), rows per page. The fields offered are those present in the context's records — if the field list comes empty, the context has no records of that entity yet (nothing uploaded, or the prompt does not emit it).
  4. Use Pré-visualizar to see the panel with real data: paging, sorting by header click, filter and search all happen on the server.
  5. Tick Publicar na galeria and Salvar. The panel now appears under Painéis (the workspace gallery) and on the workspace's Painéis tab, with the Abrir (view) and Abrir no construtor (edit) actions.

If the gallery does not show the published panel, the administrator turned off canvas-panel publishing on this installation — the builder says so next to the toggle. It is on by default.

Path 2 — through the DATTA Captain

Open SistemaDATTA Captain (or the Captain button at the top of the screen) and describe the whole goal in plain language. A complete request, with the same data as path 1:

Create the data context "Contratos" (internal name Contratos), custom category with subtype "Contratos", anchored on the Contrato entity by the contratoId property, with CPC as the rule context. Then create the "Contratos" workspace with that context and the upload and explore features, and in it a published declarative panel called "Contratos" with a table of the Contrato entity (columns contratoId, titulo and tipoContrato), source neo4j.

The Captain searches the catalog for what the platform can do, shows the plan — one step per operation, flagging those that change data and the permission required — and waits for Confirmar e executar. It also shows a preview of the panel screen before creating it.

What helps the Captain get it right the first time:

  • Name rule contexts by their internal name (CPC) or exact label; both are accepted.
  • State the anchor entity and property — without them the custom context is born without an anchor, as on the screen.
  • For a panel of a context's records, say source neo4j (the context's graph). Without it the Captain may pick a relational table that does not exist, and the panel preview comes out empty.
  • Long requests can be split: "create the context…", then "create the workspace…", then "create the panel…". Each one confirms and runs on its own.

When a step fails, the Captain says which one and why, and the following steps that depended on it do not run; those that did not depend on it may already have run (a workspace created before the context failed, for instance). Check SistemaWorkspaces and complete or delete through the screen.

To tick the context under Upload PDF and adjust prompts, use the screen (SistemaContextos) — one-click adjustments.

Final checklist

  • Context in SistemaContextos with database and index created, anchor declared (for custom) and rule corpus ticked.
  • Context ticked under Upload PDF in the visibility matrix.
  • Workspace with features, the context and (if applicable) the members.
  • Panel published, showing under Painéis with the records.
  • After the first upload: anchor entity visible in Schema do Grafo and the record's screening with text evaluated (not "no text").