For teams that move fast without giving up control.

Data Alias helps you use real customer, research, and operational data with AI — without exposing the identities behind it.

Use cases

Built for the data you work with.

  • Customer support tickets
  • Research interviews
  • Customer records
  • Employee feedback
  • Sales reports
  • Operational logs
Product

Protect identities.
Keep your data useful.

Try sample data ↗

Replace identifying details with consistent aliases, review the changes, and export your data for AI. Bring the AI response back to restore aliases that match your saved mapping.

Sample
NameEmailPlanAmount
Alex Kimalex@example.comTeam240
Alex Kimalex@example.comTeam120
Jamie Leejamie@example.comPro90
NameEmailPlanAmount
PERSON_001EMAIL_001Team240
PERSON_001EMAIL_001Team120
PERSON_002EMAIL_002Pro90

Consistent aliases. Useful data.

Replace identifying values while keeping the fields your analysis needs.

Review
  • NameAlias
  • EmailAlias
  • PhoneMask
  • API keyRemove
  • PlanKeep

Notes: name detection not run

See what changed.

Review applied rules, unscanned content, and items that need your attention.

customers.safe.csv
Download, then upload to your AI tool.
Analysis instructions
Aliases such as PERSON_001 stand for one person each. Keep them unchanged in your answer.

Ready for your AI workflow.

Prepare a protected file and clear instructions for the AI tool you use.

The round trip

Protecting the file is half the job. Getting the answer back is the other half.

The answer comes back with real names

One-shot scrubbers stop at the safe copy. Data Alias keeps the mapping — in your browser, never on a server — so when ChatGPT, Claude, or Gemini answers, you can turn PERSON_001 back into a name. Paste the answer or drop in the whole result file; anything the mapping cannot resolve is reported, not guessed.

Same customer, same alias — always

Within a file, across a batch, and across sessions with the encrypted vault, PERSON_001 stays the same person. Joins, repeat counts, and analysis structure survive the protection.

A record you can hand to a reviewer

A processing report listing counts, rules, and the file's fingerprint. It contains no values from the file.

Benchmarks

Measured in the browser.

Local processing, tested restoration, and preserved data structure.

File content uploaded during local processingNetwork audit of the local flow, Sept 2026 — no file value in 363 requests.
0 B
Third-party trackers on the audited pagesNo third-party request at all on home, pricing, security, blog and the app pages.
0
CSV rows processed in the benchmarkFive runs of a 200,000-row file; row count and column order intact.
200K
Exact-token restoration in the taxonomy benchmarkOne markdown-escaping case is a documented limitation, excluded from this count.
18 / 18
Row and column structure preservedVerified across 8,165 rows in 10 benchmark fixtures.
100%

Results apply to the tested files and local-processing flow. Optional integrations transfer data when you choose to import or save.

View benchmark details

Local processing and trackers. A scripted network audit of the production build, September 2026, watching every request made across the home, pricing, security and blog pages and the full app flow — import, review, protect, export, restore. No value from the test files appeared in any of the 363 requests, and no request went to a third-party host. The check matches known test values rather than counting bytes, and it did not exercise the Google or Notion integrations, which contact those services by design. Re-run the check yourself.

Scale. Five measured runs of a 200,000-row, 30 MB CSV through parse, protect and export in a desktop Chromium build. Output row count, column names and column order matched the input every run. Per-cell comparison was checked at the 8,165-row scale, not at 200,000.

Restoration. The exact-token tier of the restoration benchmark: 18 of 18 counted fixtures returned the original value byte-for-byte. A nineteenth fixture — an alias wrapped in markdown bold underscores — is a known limitation and is excluded from the count, not passed. Fixtures that test mutated or paraphrased aliases are scored separately and are not part of this figure.

Structure. Ten generated fixtures totalling 8,165 rows, run through the real detection and transformation engine. Row count, column count, column order, row order, untouched cells, empty cells and repeated-value alias consistency were all intact.

These are benchmark results on test files, not customer outcomes, and not a guarantee for every file. Review a safe copy before you share it.

Local by design.

Where the work happens, what is kept, and what you approve before anything leaves.

/ 01

Local processing

The file is parsed in the browser tab you have open. It is never sent to a Data Alias server.

/ 02

No file retention

Files live in tab memory only — a refresh clears the session. Saved policies and the encrypted vault save only when you choose.

/ 03

Review before export

Every detected column and protection rule is reviewable before the safe copy is created.