A result grid that keeps up
Sort, filter, pin, hide and resize columns. Search the result, open long values, or copy a row as JSON or CSV.
Open source · Made in Germany · No tracking
Write SQL. Watch it come alive.
A lightweight SQL editor and database client for the desktop. One app for PostgreSQL, MySQL, SQL Server and DuckDB.
Built for people who live in SQL.
Under the hood
- One session per tab
- Every editor tab holds its own connection. Transactions, temp tables and SET options stay where you left them. If a transaction is still open, Colwerk offers to commit or roll it back.
- Gentle on your database
- Drivers read only up to the row limit and cancel the rest. The schema tree loads on demand through its own metadata session, and table sizes come from statistics instead of COUNT(*).
- Guard rails for production
- Give a connection a color, for example red for production, and it marks the sidebar, tabs and toolbar. On request, Colwerk asks before any statement that may change data or structures.
- CSV & Parquet export
- Export the rows you see or the full result without limit. Colwerk writes the file to disk itself, so large exports never pass through the UI.
- Built-in MCP server
- Your AI assistant can browse schemas, run read-only queries and open SQL in a new tab. You decide what actually runs.
- Local by design
- Bound to 127.0.0.1 on a random port with a per-start token. Passwords are encrypted with AES-256-GCM, and the key lives in your OS keychain.
Every day
Keys in the schema tree
Primary keys, foreign keys with their target and indexes show right in the tree. Double-click a table to query all its columns.
Keyboard first
Format SQL with ⌘/Ctrl Shift F, run with ⌘/Ctrl Enter and open the execution plan with ⌘/Ctrl E. Multiple cursors and block selection included. Press ? for all shortcuts.
Sensible defaults
New tabs start with a limit of 200 rows and a 60-second timeout. Enough to see your data, too little to strain the server. Both can be changed per tab.
SQL from the schema
Generate SELECT, INSERT, UPDATE and DELETE templates with the right parameter style for each engine. UPDATE and DELETE match on the primary key.
Autocomplete and history
Autocompletion knows your tables and columns. The last 50 queries are searchable and open in a new tab with one click.
Privacy by design
Your databases. Your rules.
A database client sees everything you query. Colwerk keeps all of it on your machine.
Developed in Germany
Colwerk is built and maintained in Germany. The code is MIT licensed, so anyone can audit it.
No tracking
Colwerk has no telemetry, no analytics and no account, and it does not report what you do or which databases you connect to.
No built-in AI
Colwerk sends nothing to any AI service. If you want an assistant, you connect your own and decide which model sees your data.
Credentials stay local
Passwords are encrypted with AES-256-GCM, and the key lives in your OS keychain. They never leave your machine and are never shown to an AI.
Ask your AI. You run it.
Bring your own AI.
Colwerk has no AI inside. It speaks MCP, the open Model Context Protocol, so you can connect the cloud or local assistant you already use and decide what it sees.
- 1
Ask
“Which city buys the most coffee?” in Claude, ChatGPT or any other MCP client.
- 2
Explore
Your AI looks at schemas, tables and sample rows and runs read-only queries to find the answer.
- 3
Hand over
The final query opens in a new Colwerk tab, with an explanation as SQL comment.
- 4
You run it
Review it, change it, run it. Colwerk never runs it for you.
Your AI can
- Browse connections, schemas, tables and columns
- Look at sample rows to understand the data
- Run SELECT queries for exploratory analysis
- Open finished queries in a new editor tab
Your AI cannot
- See passwords or other credentials
- Insert, update, delete or change the schema
- Run anything in your editor tabs
- Connect without its own access token
-- Which city buys the most coffee? -- Sums coffee orders per customer city -- and returns the city with the most cups. SELECT c.city, SUM(o.quantity) AS cups FROM orders o JOIN customers c ON c.id = o.customer_id WHERE o.category = 'Coffee' GROUP BY c.city ORDER BY cups DESC LIMIT 1;
Queries from your AI run in a separate session inside a read-only transaction that is always rolled back, with a 30 s timeout and a row limit. Setup is a single entry in your MCP client; Colwerk shows it for you.
Get Colwerk
Free, with builds for macOS, Windows and Linux.