Find it.Protect it.Use it.
Classifyre finds personal data and secrets across the systems you already run, masks them automatically, and lets your team query everything with SQL, Excel and Power BI. Every read is logged. Nothing leaves your servers.
No signup. One Docker command. Your data stays with you.SELECT * FROM crm.customersPerson · 0.97nameredacted | IBAN · 0.99ibanlast four | Email · 0.98emailhashed | Secret · 0.96api_tokenhidden |
|---|---|---|---|
| Anna Bergerredacted | AT61 1904 3002 3457 3201••••3201 | anna@example.at5fc7613cacc41469 | tok_7Qe2x9 |
| Lukas Hoferredacted | AT48 3200 0000 1234 5864••••5864 | lukas@example.at8169dd18e1635846 | tok_m81Kp3 |
| Marie Novakredacted | DE89 3704 0044 0532 0130 00••••3000 | marie@example.de8a84971c7a971d8a | tok_x0Pd5a |
| Jonas Weberredacted | AT02 2011 1000 0123 4567••••4567 | jonas@example.org2f122056582e0a94 | tok_Zt4w8c |
1 column hidden. Personal columns are masked inside the query. The token column is hidden.
- ConnectIceberg tables via S3
- DetectAWS key found · 0.96
- GovernSecret hidden
- GrantRequest approved
- Shareorders + customers
- Connect BIExcel · own login
- LogAgent read · MCP
docker run -d --name classifyre \ -p 3000:3000 \ --shm-size=1g \ -v classifyre-pgdata:/var/lib/postgresql/data \ -v classifyre-data:/var/lib/classifyre \ -v classifyre-uv-cache:/cache/uv \ classifyre/all-in-one:0.9.0Your data is useful. It is also full of personal data.
Every team wants the numbers. Every table that holds them also holds names, accounts, addresses and the odd password. So access becomes a negotiation, and the negotiation ends in a spreadsheet.

- Field note 01
Who can see this column?
Finance asks for the customer database. Which of its 212 tables hold personal data? Nobody knows, so the ticket waits, or everyone gets everything.
- Field note 02
Just send me the export
Data leaves by CSV and shared database passwords. Six copies later nobody can say where the customer list lives, or who still has it.
- Field note 03
Prove it to the auditor
Who read the payroll tables last quarter, with which tool? Without a log of every read, the honest answer is: we don't know.
It is the same shape every time: useful data locked away because nobody can say what is personal in it, or handed out whole because nobody had time to say no. Masked by default, granted on purpose, every read on record: that is the way out.
From a raw table to a shared dataset.
Follow the data through Classifyre: found where it lives, queried in place, governed, used in the tools your team already has, masked for the people who should not see it, and shared on purpose.
- 01Connect & detect
Find personal data and secrets where they live
Connect databases, lakehouses, buckets and SaaS tools; Classifyre reads them in place, so nothing is copied into another warehouse. Built-in detectors look for 41 kinds of personal data, including Austrian, German and Swiss identifiers, and 27 kinds of secrets. Add your own with a regex, a model or a prompt.
Every finding comes with a score and a written reason, so a data protection officer can check it and a security team can rank the cleanup. The classification it produces is what the masks follow later.
- 40+ source types
- 41 PII entities · 27 secret patterns
- A score and a reason on every finding
Scan · 3 sources- crmPostgreSQLScanned
- wikiConfluenceScanned
- exportsParquet in S3Scanned
Findings, ranked- IBANcrm.customers.iban · Valid checksum in 4,211 of 4,215 rows0.99
- AWS access keywiki / Deploy notes · Key and secret on the same page0.96
- AT SVNRexports/2026-10.parquet · Check digit valid, column named svnr0.93
- Emailcrm.tickets.body · Addresses in free text, 312 tickets0.81
crm.customers.iban classified as Personal · masked from the next query
Personal data found across three sources, ranked with reasons. - 02Query across sources
Query every system with one SQL statement
Ask questions across the systems you already have: PostgreSQL, MySQL, Databricks, and Delta, Iceberg and Parquet files in S3, Azure or Google Cloud Storage. Join up to eight sources in one query. Each query runs next to the data; nothing is loaded into a central store first.
Results go where people work: the SQL page, exports to Excel, CSV and Parquet up to a million rows, or straight into a BI tool.
- Up to 8 sources in one query
- Delta · Iceberg · Parquet · CSV
- Exports up to 1M rows
Query datacrm · PostgreSQLlake · Parquet in S3▶ RunSELECT c.email, c.city, sum(o.amount) AS revenueFROM crm.customers cJOIN lake.orders o ON o.customer_id = c.idGROUP BY 1, 2;3 rows · 2 sources joined · email hashedemail city revenue 5fc7613cacc41469 Graz 4,812.40 8169dd18e1635846 Linz 1,130.00 8a84971c7a971d8a Wien 2,977.15 A PostgreSQL table joined with Parquet files in a bucket. - 03Governance
Governance: an owner, a contract and a history for every table
A governed table has an owner, stewards and a custodian, a classification for every column, and a status from draft to certified. Quality rules check it on a schedule: missing values, invalid IBANs, duplicates, row counts or any SQL you write. The agreement exports as an ODCS data contract.
Usage shows who reads the table and through which tool; Activity shows every read and every change, old value next to new. When the auditor asks, the answer is already written down.
- Owners, stewards, custodians
- Quality rules on a schedule
- ODCS data contracts
- Every read and change on record
crm.customers · GovernanceCertified5 of 5 done✓ CertifiedOwnership
- OwnerGroupFinance
- StewardGroupData office
- CustodianGroupIT platform
Columns
Column Classification Access iban Personal data Masked · shown as last four email Personal data Masked · shown as hashed api_token Secret Hidden Quality
+ Add rule- email · Missing values = 0
- Row count ≥ 4,000
- Check a column · iban · Invalid values ≤ 0.5 %
Run now0.1 % invalid · passedUsage · last 30 days
- Reads
- 1,284
- Exports
- 37
- Readers
- 14
Through: SQL connections, the app, MCP
↓ Export data contract apiVersion: v3.2.0kind: DataContractstatus: activeteam: [Finance, Data office]schema:- name: customersproperties:- name: ibanclassification: personalquality:- type: librarymetric: invalidValuesmustBeLessThan: 0.5Owner, steward and custodian assigned, access set, a quality rule run, the table certified, the contract exported. - 04Power BI & Excel
Connect Power BI and Excel without a shared password
Classifyre answers as a PostgreSQL database. Each person creates their own SQL connection, encrypted and expiring, and points Power BI, Excel, Tableau or DBeaver at it. No service account with the keys to everything, no password inside a report file.
The connection carries the person's own rights: the tables they may query, the rows their group may see, the masks on every personal column. Revoke it and the report stops.
- A personal connection per tool
- TLS only, expiring, revocable
- The read log names the tool
SQL connections · New connection- Name
- Power BI
- Tables
- crm.customers, sales.orders
- Expires
- in 90 days
CreatedServer=sql.example.at:5432;Database=finance;User=lena_power_bi;SSL=verify-fullCustomers report · via SQL connection Power BIGrazLinzWienSalzburgredacted ••••3201 1,240.00 redacted ••••5864 312.50 redacted ••••3000 88.90 09:14 · Lena read crm.customers via SQL connection Power BI
A personal connection, used from a BI report, still masked. - 05Mask per person
Masked for one person, clear for another
Personal columns are masked by default: redacted, last four, hashed so tables still join, or hidden. Masks apply inside the query, so a WHERE or a JOIN can't reach past them. Only a group granted Data reader, unmasked reads the clear values, for a source or a single table, with an end date. Admins don't get it by default.
The hashed mask is pseudonymisation in the sense of the GDPR, not anonymisation: equal values hash the same, so analysts can join and count people without reading who they are. Watch it below: the same queries, run by an analyst and by a finance steward.
- Masked by default
- Redacted · last four · hashed · hidden
- Pseudonymisation, not anonymisation
- Clear reading is its own grant
▶ RunSELECT city, count(DISTINCT email) AS customers FROM crm.customers GROUP BY city;Lena, analyst
Data reader
4 rowscity customers Graz 1 Linz 1 Wien 1 Salzburg 1 Tom, finance steward
Data reader, unmasked
4 rowscity customers Graz 1 Linz 1 Wien 1 Salzburg 1 Counts work on masked columns. The answer is useful; the person behind it stays covered.
Reading masked data is a grant of its own, with an end date, and it is on record.
This is the part most tools leave to you.
- EMLThe external email itself: headers, recipient, timestamp
- IDSender identity resolved across the directory
- CLASSClassification marking found inside the attachment
- DUPInternal original confirmed by fingerprint
The data looks different. The rules are the same.
Classifyre reads whatever the organisation runs on; the sector only decides which columns are personal and who reads them in clear.
- Public sector · Länder · citiesTypical dataRegisters, case files, statistics, procurementWhat changesDistricts query their own rows, personal columns arrive masked, the DPO reads the log
- Healthcare · hospitalsTypical dataAdmissions, lab results, staffing, billingWhat changesAnalysts count patients without reading who they are
- Banking · KYCTypical dataCustomer master, transactions, accountsWhat changesIBANs as last four, hashed IDs that still join, a read log for the regulator
- InsuranceTypical dataClaims, policies, health declarationsWhat changesActuaries get the tables, not the policyholders
- BI · analytics teamsTypical dataWarehouses, lakehouses, Delta and Iceberg tablesWhat changesPower BI and Excel on personal connections instead of one shared password
- Security teamsTypical dataGit, wikis, tickets, buckets, logsWhat changesKeys and tokens found, ranked with reasons, cleaned up first
- Internal audit · complianceTypical dataAccess logs, payroll, contractsWhat changesWho read which table, through which tool, and when: one filter
- Incident responseTypical dataMail, drives, databasesWhat changesOne fingerprint across systems, worked as a case
Each row is a starting point, not a separate product: the same masks, grants and read log underneath.
Open your data safely tonight.
Point it at one database you already run. By the morning you know which columns hold personal data, and your team can query the rest. Everything you set up carries over when you go to Kubernetes.
One command tonight. The same core at scale.
The all-in-one image has the database, the UI and the scan workers in it; everything stays on your machine. The Helm chart runs the same core on Kubernetes when the estate grows.
docker run -d --name classifyre \ -p 3000:3000 \ --shm-size=1g \ -v classifyre-pgdata:/var/lib/postgresql/data \ -v classifyre-data:/var/lib/classifyre \ -v classifyre-uv-cache:/cache/uv \ classifyre/all-in-one:0.9.0Or run it on Kubernetes
Ephemeral scan and query workers scale to zero between runs and fan out as far as your estate goes. Your cluster, your data.
helm install classifyre \
oci://registry-1.docker.io/classifyre/classifyre-core \
--version 0.9.0- Tonight, on your machine
One Docker command. Sources, masks and the read log stay local.
- At scale, on Kubernetes
The same core as a Helm chart: workers scale to zero between runs.
- When it becomes infrastructure
Your team or your integrator runs it. Enterprise tailors it to your organisation, adds OpenShift and an SLA. Every feature is already open source.


