Drive (GrafDrive)

GrafDrive turns a folder of scattered, duplicated corporate files into a harmonized, queryable knowledge graph. It crawls files, extracts metadata, detects duplicate/overlapping datasets, proposes relationships, and lets you review everything before producing a cited harmonization report.

Find it under Drive (/drive).

The workflow

Upload folder → Crawl & profile → Detect overlaps → Review proposals → Export report

Drive is read-only and human-reviewed — it never mutates your source files, and nothing enters the graph without your approval.

Workspaces

A workspace is one harmonization project. Create a workspace, then upload a folder into it. The three-pane layout is:

┌────────────┬──────────────────────┬──────────────┐
│  Workspace │   Tab content        │  Context     │
│  sidebar   │   (Overview, Files,  │  drawer      │
│  (260px)   │    Graph, Review …)  │  (360px)     │
└────────────┴──────────────────────┴──────────────┘

The context drawer opens when you select an artifact or a proposal.

Tabs

Tab	Purpose
Overview	Workspace summary — file counts, sources, status
Files	Table of crawled artifacts (name, rows, owner, columns)
Graph	Visual graph of datasets and their detected relationships
Ontology	The schema Drive is harmonizing toward
Schemas	Side-by-side column comparison across datasets
Review	Queue of proposed relationships to approve/reject
Ask	Natural-language questions over the workspace graph
Reports	Generate and download the harmonization report

Reviewing proposals

The Review tab is the human-in-the-loop step. Each candidate relationship shows its type, a confidence score, an explanation, and supporting evidence.

Proposals are color-coded by status:

• 🔵 Proposed — awaiting your decision
• 🟢 Approved — accepted into the graph
• 🔴 Rejected — dismissed
• 🟠 Uncertain — flagged for a closer look

Filter by status, type, or minimum confidence. Click a proposal to open the evidence panel and see exactly why it was suggested before approving.

Exports

From the Reports tab you can generate a harmonization report and download it. Formats include Markdown (.md) and JSON (.json) — a cited summary of datasets analyzed, duplicates found, canonical sources chosen, and the resulting graph structure.

Example

A typical scenario: 47 supplier files across 4 regional folders (supplier_master_v3.xlsx, vendors_2024_final_FINAL.csv, Copy of suppliers (2).xlsx…). Drive:

1. Crawls all 47 files and profiles their columns.
2. Detects that supplier_id ↔ vendor_code and company_name ↔ vendor_name map across datasets (with confidence scores).
3. Flags likely duplicates for review.
4. After you approve, produces a canonical supplier_master with owners attributed and a queryable, traceable graph.

See the marketing overview at graffold.com/graph-drive for the pitch version.