User guide

Everything you have to do, step by step — from first upload to delivered data.

  1. STEP 1

    Upload your files

    • Go to the Extract page and drop your files onto the upload zone — or click "Browse files".
    • Supported formats: .xlsx, .xls, .csv and .pdf.
    • Desktop Professional only: click "Browse folder" to queue every supported file in a folder at once.
    • Messy files are fine — branding rows, empty rows and multi-row headers are detected automatically.
    • No files handy? Download one from "Try it with sample files" further down this page.

  2. STEP 2

    Choose what to extract

    • "All fields" (default): the full 43-column canonical PI schema.
    • "Custom": type the fields you need (e.g. Employee Name, DOB, SSN) — one per line or comma-separated — then click "Match fields". Green chips = recognized, red = unknown.
    • "Template": pick a saved template, or click "New from Excel" and upload a workbook whose headers ARE the fields you want.
    • Save any custom field set as a template so you never type it again.

  3. STEP 3

    Process

    • Click "Process files". The first time, accept the privacy agreement (one click).
    • Watch the log panel: header detection → column mapping (cache / fuzzy / embedding / LLM) → row scoring.
    • Cloud: 1 successful document = 1 credit. Failed files cost nothing.
    • Desktop Professional: batches run 3 files in parallel, unlimited.

  4. STEP 4

    Review the results

    • Every file gets a verdict — ACCEPT (ready to use), REVIEW (needs your check) or REJECT.
    • Files marked REVIEW appear in the Review queue. Open "Show details" → "Column Mapping".
    • Wrong mapping? Change it in the dropdown — your correction is learned and applied to all future files automatically.
    • Click "Approve and use" or "Reject" to finish the review. Decisions are saved.

  5. STEP 5

    Export

    • "Download spreadsheet" gives you the extracted Excel immediately.
    • "Export as" offers CSV, Excel, DAT and JSON for one-off downloads.
    • For the full business workflow, use the Delivery Center (next step).

  6. STEP 6

    Deliver (Desktop Professional)

    • Open the Delivery page and select your accepted files.
    • Stage 1 — Prepare: deduplicate & clean the Excel output.
    • Stage 2 — Convert: generate the §-delimited CSV built for fast bulk loading.
    • Stage 3 — Load: open the SQL wizard, pick MySQL/PostgreSQL/SQLite/SQL Server/Oracle, test the connection, upload. Credentials are used once and never stored.
    • Stage 4 — Deliver: generate the tab-delimited .dat delivery file.

  7. STEP 7

    Tune it (optional)

    • Settings → Models: choose the local LLM (default qwen3:8b) and embedding model.
    • Settings → Thresholds: control how strict matching and row quality checks are. Defaults work well — change them only if you know why.
    • Settings → Alias Cache: inspect everything the system has learned; export it as JSON or clear it.

Standard operating procedure

1

Open CyberInci and wait for the splash to finish

First launch of the day takes ~1 minute while the engine warms up. The workspace always starts empty — results are per-session.

2

Extract: drop your files (or Browse folder), pick fields, click Process

Failed files never cost anything. Everything is processed on your machine (Desktop).

3

Review: clear the Review queue

Fix any wrong column in the dropdown — the fix is learned and applied to every future file.

4

Delivery: select the accepted files and run the stages in order

1 Prepare (dedupe Excel) → 2 Convert (§-CSV) → 3 Load (SQL wizard) → 4 Deliver (.dat).

5

Collect the outputs

Download links appear under each stage; SQL rows land directly in your database.

Try it with sample files

No files handy? Download a sample, drop it on the Extract page and run the whole SOP end to end. All sample data is fictional.

Messy Excel export

Branding rows, a blank row, cryptic headers (EMP_NO, SOC_SEC…) and one duplicate row — watch header detection skip the junk and Delivery Stage 1 remove the duplicate.

CSV with wordy headers

Human-style headers like "Social Sec #" and "Date Birth" — watch the fuzzy and embedding tiers map them to the canonical schema.

What can CyberInci extract?

These are the file shapes the engine handles — every preview is fictional sample data.

quarterly_roster.xlsxxlsx
GLOBEX INC — INTERNAL USE
EMP_NOEMPL_NMSOC_SEC
1001DOE, JANE B000-12-3456
1002SMITH, ALEX J000-23-4567

Header found on row 3 — branding skipped

Messy Excel exports

Branding rows, blank rows and logos above the real table — the header is found automatically, wherever it hides.

regional_offices.xlsxls
EastWestCentral

3 sheets detected — all extracted

Multi-sheet workbooks

Every sheet is processed, scored and combined — no copy-pasting tabs together first.

contact_list.csvcsv

Employee Number,Name of Employee,Social Sec #,Date Birth

1001,DOE JANE B,000-12-3456,01/14/1990

"Social Sec #" → SSN"Date Birth" → DOB

CSVs with human headers

Wordy or abbreviated headers are matched to the canonical schema by the fuzzy and AI tiers — no renaming needed.

benefits_report.pdfpdf
NAMEIDDOB
DOE, J100101/14/90
SMITH, A100207/22/85

1 table found on page 2

PDFs with tables

Tables inside PDF reports are detected and extracted directly — same pipeline, same confidence scores.

old_records_scan.pdfpdf
OCR

Needs the free Tesseract OCR add-on

Scanned PDFs (OCR)

Image-only scans are read with the optional OCR add-on, then extracted like any other document.

export_no_headers.csvcsv
Employee ID?Name?SSN?
1001DOE, JANE B000-12-3456
1002SMITH, ALEX J000-23-4567

No header row — columns inferred from values

Files with no headers at all

When there is no header row, columns are inferred from the data itself — IDs, names, SSNs and dates are recognized by shape.

Troubleshooting

"Network error" when processing

The extraction engine isn't reachable. Desktop: close and reopen CyberInci — the engine starts automatically (the very first start can take about a minute). Cloud: the service may be restarting — try again in a minute.

Desktop app says "Engine failed to start"

Restart the app — that resolves most cases. If it happens again, your antivirus may be blocking the engine: add CyberInci to its exclusions and restart. Still stuck? Send support the log file at %LOCALAPPDATA%\CyberInci\logs\engine.log — it contains the exact error.

AI assist (Tier 4) always shows 0 mappings

The optional local AI model isn't installed. Install Ollama, then run `ollama pull qwen3:8b` once. Extraction works fine without it — tiers 1-3 handle most files.

Scanned PDF fails

Reading scanned (image-only) PDFs requires the free Tesseract OCR add-on. Install it, restart CyberInci, and process the file again. PDFs with selectable text work without it.

Out of credits (Cloud)

Buy a credit pack from the coins badge in the navbar, or switch to Desktop Professional for unlimited local processing.

I have thousands of files (3,000–10,000) in a folder

The in-app folder picker is for smaller, interactive batches. For very large folders, use the bundled batch tool: `python batch_process.py C:\your\folder -o C:\output`. It processes files in parallel, shows live progress (done/total, rate, ETA), writes a manifest.csv of every result, and files outputs under accept/review/reject. Roughly 3,000 files ≈ 30 min, 10,000 ≈ ~1.5 hours.

Ready to try it?

Unlimited local processing — start extracting.