mask mode
Pseudonymization, in place.
Read real rows, replace PII with realistic fakes, generalize quasi-identifiers. Keeps your exact row counts and joins.
✓Replace PII in-place
✓Generalize quasi-IDs
✓k-anonymity + l-diversity check
Clone production data safely.
Point cloneside at your database. It classifies every PII column, then either masks it in place or generates a synthetic clone — referentially intact, safe to hand to Claude Code, Copilot, or your team.
the problem
Developers are already pasting production data into AI.
Copying a real table to debug a query. Dropping a CSV into Claude to write a migration. Sharing a dump so Copilot has context. It happens dozens of times a day — and every time, real names, emails, and IDs leave your control.
⚠PII in prompt history
⚠Data in model context
⚠Compliance exposure
users · productionreal
a.chen@acme.ioPII
Ada ChenPII
+1 555 0142PII
94110QUASI
users · clonesafe
u_8f3@example.netSAFE
Quinn DiazSAFE
+1 555 0000SAFE
941xxSAFE
One scan. Two ways to stay safe.
cloneside shares the scan and classify stages, then diverges where it counts — strategy and emit.
synthetic mode
Statistical generation.
Learn the distributions, then generate brand-new rows that look real and carry zero original records.
✓Sample from distributions
✓Conditional sampling
✓Distribution parity check
how it works
From connection string to safe clone.
Three shared steps map your data. One fork decides how it's made safe. Both paths converge on a single, referentially-intact output.
01
Point at your database
One connection string — Postgres, MySQL, or a dump file.
postgresql://prod/db
→
02
Introspect the schema
Walk every table, column, and foreign key — no config.
→
03
Classify every column
Each field tagged PII, SENSITIVE, or SAFE — with a confidence score.
PIIQUASISAFE
↓ choose a strategy
mask mode
Pseudonymize in place
Replace every PII value with a realistic fake and generalize quasi-identifiers. Same row counts, same joins — verified safe.
replace PIIgeneralize quasi-IDsk-anon + l-diversity
synthetic mode
Generate from scratch
Learn the distributions, then sample brand-new rows that look real and carry zero original records — parity-checked.
sample distributionsconditional samplingdistribution parity
Emit a safe clone
A referentially-intact database — same shape, same joins, zero real records. Ready to ship anywhere.
✓ clone written
use safely with AI
Hand it to your coding agents
Point your AI tools at the clone instead of production — real-shaped data, full referential integrity, zero leak risk.
Claude Code
Copilot
Codex
The cloneside CLI runs entirely on your machine — your credentials and data never leave it.
Hosted scan mode is available for non-sensitive databases and is labeled as such.
Pricing — early access
Simple tiers. No checkout yet.
Tell us which tier fits — we'll reach out when it opens.
Free
$0
Open source · Apache-2.0 · never calls home
- ✓cloneside scan — full PII taxonomy
- ✓cloneside clone — Faker fidelity
- ✓Referential integrity + parity suite
- ✓MCP scan tool for Claude Code / Copilot
- ✓--json output for shell AI pipelines
- ✓PostgreSQL + MySQL
- ✓One user · unlimited runs
pip install cloneside
Pro
$49/mo
Everything in Free, plus:
- ✓Query gateway + AST firewall
- ✓Schema-alias vault (opaque + semantic)
- ✓Statistical-fidelity clone
- ✓ε-Differential privacy (configurable)
- ✓Mask mode — k-anon + l-diversity
- ✓HIPAA / GDPR / LGPD audit reports
- ✓Pro support
For regulated teams
Team$119 / user / yr
Shared policies · dashboard · audit logs · priority support
Compliancefrom $1,500 / mo
HIPAA / LGPD / PIPEDA evidence packs · SSO · DPA · deployment support
Contact us →Using SQL Server? We're adding it.