Karg← HomeDocs
What Karg does with your data
What you can upload, what Karg changes, what it leaves untouched, and what you can export. Start with the quick answers, then read the details below.
What will Karg change?
It standardizes formats, fixes clear typos, removes or merges duplicates, and flags anomalies.
See what changes →What will Karg not change?
Your original file is preserved. Karg never invents or guesses missing data; anything it is unsure about goes to human review.
See how review works →What can I export?
Cleaned data as your original format, CSV, XLSX, or JSON, plus a full audit log.
See exports →01 · File types
Supported file types
Karg reads common tabular and structured data files. It works best on data with a clear header row and columns.
02 · Detection
What Karg detects and fixes
Karg profiles every column and proposes a fix for each problem it finds. Deterministic rules run first, and AI is used only where the data is genuinely ambiguous.
- Inconsistent formats in dates, phone numbers, email addresses, and addresses.
- Empty or placeholder values converted to a true null.
- Duplicate records detected and flagged for your review, rather than merged silently.
- Inconsistent values standardized to one form.
- Anomalies flagged for a closer look.
- Column types converted where a column holds the wrong type.
03 · Review
How review works
Karg routes each proposed change through confidence-gated tiers (rules first, then ML, then AI) and only changes what it is confident about.
- Anything below the confidence threshold is sent to a human review queue instead of being changed silently.
- Each suggestion carries a confidence level: high (90% and above), medium (70 to 90%), or low (below 70%).
- You approve or reject each suggestion, and only approved changes are applied.
04 · Exports
How exports work
When you are done, download your cleaned dataset with only the changes you approved.
- Export as your original format, CSV, XLSX, or JSON.
- The download reflects approved changes only.
- The audit log can be downloaded alongside the data.
05 · Audit log
How audit logs work
Every change Karg makes is recorded so you can verify each decision.
- Each entry records the column, what changed, why, the confidence, and the source (a rule, ML, or AI).
- The audit log is downloadable as a separate file.
- Items sent to review are recorded too, including your approve or reject decision.
06 · Security
Security basics
A few basics about how your data is handled.
- Access requires signing in.
- Each organization's data is isolated from others, enforced by database row-level security.
- Your original upload is preserved; Karg works on a copy.
07 · Limitations
Known limitations
What Karg is not, so you know what to expect.
- It works best on tabular data with clear columns; deeply nested or free-form documents are not its focus.
- It does not enrich or invent missing data; uncertain values are escalated for review rather than filled with a guess.
- A human should review escalated items before relying on the cleaned output.
- Results depend on the cleaning profile you choose, from faster to more thorough.
See Karg run on your own data.
Upload a messy dataset and watch it get cleaned, with every change shown.
Request early access