Work with information — Docs

Start with Qora

Configure your model provider, upload a document, ask about it, and inspect the saved information.

1

Configure your model provider

Your Qorbit subscription provides the shared database and its tools. Choose Anthropic / Claude, OpenAI, xAI / Grok, Moonshot AI / Kimi, or DeepSeek. The API key authorizes the provider, and the model you select is what runs. That provider and model power Qora and classification. The model must support tool calls and images. Model-provider usage is billed separately. Follow the setup on Connect. This key is not the authorization for an external AI client.

2

Upload a document

In Qora, select + → Add files. See Upload documents through Qora for accepted types.

3

Ask about it

In Qora, ask about the uploaded document. Qora uses your configured key for the conversation.

4

Inspect the saved information

Open Neural Fabric and confirm the entity was saved. See Search and inspect.

Upload documents through Qora

Upload documents through Qora, or attach a source file through fabric_add. Qorbit processes the information for use in Neural Fabric.

In Qora, select + → Add files.

Save and test the provider key on Connect before uploading. The same saved provider and model power Qora and classification, including images. If the model list cannot load, enter an exact model ID. If an upload fails, review the reported error before retrying.

Attaching a file to an external client’s chat does not automatically transfer it to Qorbit. Upload it through Qora or submit it through fabric_add with sourceDocument.

Qora accepts PDF, plain text, Markdown, CSV, JSON, XML, YAML, DOCX, and XLSX.

.pdf .txt .md .csv .json .xml .yaml .docx .xlsx

After processing, confirm that the entity appears in Neural Fabric. An uploaded file alone does not confirm that an entity was saved.

Add structured entities

Call fabric_add with entityType and properties. Properties are classified before storage. To express a relationship, include the other entity’s identifying value as a property. You cannot create relationships manually.

fabric_add · sample

{
  "entityType": "SalesOrder",
  "name": "SO-03242607",
  "properties": {
    "orderNumber": "SO-03242607",
    "customerName": "Clearview Commercial Lighting",
    "quoteReference": "QT-03242607",
    "customerPo": "CCL-2026-R117",
    "orderDate": "2026-04-02",
    "amount": 510105,
    "status": "open"
  },
  "actor": "docs-sample"
}

Optional: entityId, name, textContent (short searchable annotations, not document bodies), mode (upsert, create, update), idempotencyKey, layer, actor (a caller-supplied provenance label, not verified identity). Keep 60 or fewer top-level keys and string values under 3,500 characters. A provider API key is required before the first write.

Attach a source file with `fabric_add`

A separate path from Qora upload: include sourceDocument with base64Content, filename, and mimeType. PDF processing over MCP requires processPdfOverMcp: true; prefer Qora’s + → Add files for PDFs.

The original file can be stored as an attached source. Use fabric_read_source to read its extracted text (default 100,000 characters per call, maximum 200,000).

Search and inspect

Use fabric_search for questions or text search. It supports match and answer. When mode is omitted, search defaults to match. Answer generation requires explicit mode: "answer" and uses a model to compose a response from stored information. Use fabric_list for inventory and full entities. Use fabric_read_source when an entity has attached source documents or truncated content.

In the app, open Neural Fabric, search, and select an entity to inspect properties. Deep link: /app/graph?entity=YOUR_ENTITY_ID.

Explore connections

Open an entity in Neural Fabric to see connected entities and why they connect. From MCP, call fabric_graph with includeExplanations: true. Edges are harmonic connections derived from shared properties. Modes: root, tenant, resonance. Planes: truth, resonance, full, provenance.

Update entities

fabric_edit requires entityId and edits (1–25). Supported operations: set, remove, rename, setName, setSemanticRole, removeSemanticRole. Set uses a JSON pointer such as /properties/status. Ordinary edits do not require classification; reclassification is optional.

fabric_edit

{
  "entityId": "YOUR_SALES_ORDER_ENTITY_ID",
  "edits": [
    {
      "op": "set",
      "path": "/properties/status",
      "value": "released"
    }
  ],
  "reason": "Release sample sales order",
  "actor": "docs-sample"
}

Inspect history

fabric_provenance requires entityId or relationId. The trail is append-only. Responses include entries with actor, timestamp, and operation, plus verification.status. actor is the caller-supplied provenance label from the write; it is not verified identity. verification reports trail integrity. Observed operations include entity_create, entity_update, entity_edit, and entity_delete. An entity_edit entry can include editDetails.fieldDiffs with path, before, and after.

Delete information

fabric_delete permanently removes an entity and its attached source documents after writing delete provenance. Provide entityId; optionally include reason and actor (a caller-supplied provenance label, not verified identity).

fabric_delete

{
  "entityId": "YOUR_SALES_ORDER_ENTITY_ID",
  "reason": "Remove sample data",
  "actor": "docs-sample"
}

After deletion, fabric_provenance with the same entityId still returns the recorded history, including the entity_delete entry. fabric_list still accepts includeDeleted as an inventory filter. That filter does not restore or return a hard-deleted entity.

Sample sales order

Add the sample with the quickstart. The compact Find → Explore → Update → History demonstration is on the Developers page.