The Confluence page is from two years ago.
Columns get added and meanings drift, and nobody remembers to go back and edit the page, so new hires end up learning the schema the way everyone before them did, by interrupting whoever looks least busy.
Private database workspace · on-premise agent
Living database documentation for private databases.
Taavik writes the documentation nobody has time to write, then rewrites it on every scan. An agent runs inside your network and reads the catalog directly, so what lands in the workspace is the database as it stands this morning rather than the version somebody described in a wiki two years ago and never went back to. Your credentials stay where they are.
Explicit query results only
Read-oriented guardrails
PostgreSQL, SQL Server, MySQL, Redshift
E2E encrypted
Cloud sees ciphertext, not rows
Team SQL
Saved, parameterized, audited
EU hosted
Control plane in the EU. DPA on request
The problem
You already know where the knowledge is. That is the problem.
It lives in three heads, a Confluence page somebody wrote in a hurry, and a folder of .sql files on one laptop. No two of those agree, and the only way to settle an argument is to ask whoever has been here longest, which works right up until the week they are away.
All three symptoms have the same shape. The knowledge exists, it is just not anywhere a second person can reach it.
Operational SQL lives in pgAdmin tabs and Slack DMs.
The query that untangled last month's incident sits on a laptop, the reconciliation one is in a private gist, and neither is anywhere the next person would think to look.
Finance asks for the same CSV every Monday at 8:30.
Somebody opens the query, exports to Excel, attaches the file and hits send, week after week, until the week they are away and the report simply does not go out.
And when you ask an assistant
The assistant guesses a value, and the query lies.
It writes WHERE status = 'completed' and your data says COMPLETED. The query runs. It raises nothing. It returns zero rows, and whoever reads the answer concludes there are no completed orders. An assistant that has never seen your database guesses names and guesses values, and the guesses that do not crash are the ones that cost you.
Taavik hands it the real catalog instead, through an MCP connector your assistant adds once: your tables, the relations your database never declared, and the values a column actually holds. No credentials change hands, and schema access is on the free plan.
Already using MCP? This is the address
https://api.taavik.com/mcp
What you get
Three things, one workspace.
01
Documentation
Documentation that follows your schema.
Taavik scans your database schema on a schedule and keeps an internal Markdown documentation layer aligned with the latest structure. When tables or columns are added or modified, the documentation is updated and the change history remains visible. Manual notes you wrote stay where you put them.
- ·Auto-generated after the first scan
- ·Updated automatically when the schema changes
- ·Manual notes preserved across updates
- ·Change history per object
02
Shared queries
Save the SQL your team actually uses.
Turn recurring operational SQL into shared, reusable queries. Organize by folder, bind to a connection, parameterize with variables, run safely from the app through the on-premise agent. Stop losing queries in Slack, gists, and laptops.
- ·Folders, descriptions, team and personal scope
- ·Variables for the bits that change between runs
- ·Read-only execution through the agent
- ·Audit log per run
03
Schedules
Schedule the exports your team already runs by hand.
Run a saved query on a schedule and deliver the result as CSV or Excel by email or Slack. Useful for daily checks, finance extracts, reconciliation lists, compliance support, and any recurring operational workflow that lives in someone's calendar reminder.
- ·Hourly, daily, weekly cadence
- ·CSV or Excel output
- ·Email or Slack delivery
- ·Run history and failure handling
Built for private databases
The catalog is in the cloud.
Your data is not.
Taavik uses an on-premise agent to reach databases that should not be exposed to the public internet. Query execution and scheduled exports run through the agent. The cloud acts as a control plane for configuration, documentation, and delivery.
Read the security modelSealed credentials
The browser seals the database password with X25519. The cloud holds only the envelope.
Outbound only
The agent opens TLS towards the cloud and listens on nothing, so the only rule anyone has to approve is the outbound 443 that is already open.
Read-oriented execution
Point the connection at a read-only role. Anything that writes is rejected before it is dispatched, and the agent enforces its own timeout on whatever survives.
Audit on every run
Rendered SQL, variables, who triggered it, when, on which connection, row count.
No row data in cloud DB
Catalog metadata and scheduled export results live in encrypted storage; rows are not persisted in our database.
EU hosting
Control plane runs on dedicated VMs in the European Union. DPA on request.
Guides
Start with the operational SQL workflows teams search for.
Practical guides for scheduled SQL exports, private database execution, and Slack or email delivery.
Email exports
Schedule SQL queries to CSV email
Save a recurring query, generate CSV, and deliver a secure download link by email.
Private databasesScheduled SQL exports for private databases
Run recurring exports through the on-premise agent without direct cloud database access.
OperationsSQL query scheduler for operations teams
Replace calendar reminders and scripts with visible scheduled query runs.
Get started
Start documenting your database in under ten minutes.
Install the agent on a host inside your network, seal the database credentials in the browser, and run the first scan. The documentation is waiting when it finishes.