Changelog
What shipped, and when.
Dated, specific, and written down the day it happened. If an entry says a behaviour changed, it changed.
-
Querium Pro, an AI assistant, and a monitoring dashboard
The first version you can buy. Pro is a one-time payment that keeps working forever, with a year of updates - and everything that was free stays free.
- Querium Pro: buy from the app or the site, and the key activates with one click from the email or the thank-you page
- An AI assistant that uses your own key with Anthropic, OpenAI, Google, Ollama or any compatible server - it sees table and column names, never your rows, and shows you the request before the first one is sent
- Ask, explain a query, suggest a faster one, or explain an error - every statement it writes is yours to insert, never run on its own
- A server dashboard: live activity, the tree of who is blocking whom, and table and index health
- Paste a whole connection string, duplicate a connection, and SSH through your agent and your ~/.ssh/config
- Usage statistics and crash reports, off until you say yes, with the exact report shown before anything is sent
-
Sorting on the server
Sorting used to apply to the rows on screen. When those are ten thousand of forty million, "the ten most recent" is a wrong answer in the shape of a right one. Now the real answer is one click away, in the warning itself.
- Sort a truncated result on the server rather than the page
- Once the database is doing the sorting it keeps doing it - a second click cannot quietly go back to sorting one page
- Empty values sort to the same place whether the app or the database did it
- Anything that writes is refused rather than silently run a second time
-
The row limit became the server’s problem
Querium used to ask for the whole result and throw away everything past the limit - forty million rows across the network to show you ten thousand of them. Now it asks for exactly what it needs.
- Rows past the limit are never computed and never sent
- The “there is more” warning is now exact rather than a guess
- Anything that cannot be handled this way keeps the old, slower, always-correct path
-
Export to a file, without holding it
Exporting used to mean copying, which meant the result that most needed to leave was the one that could not. Now it writes straight to a file, at any size.
- CSV, TSV, JSON, Markdown and INSERT statements, written straight to disk
- Export what is on screen, filters and sort included - or re-run and write every row
- Anything that writes is refused rather than run a second time to fill a file
-
Connections that reach production
Every TLS mode your server offers, SSH tunnels that really do check the host key, and passwords that stay in your system vault.
- Encrypting the connection and verifying the server are two settings, not one
- verify-full really checks the server’s name, tunnel or no tunnel
- An unrecorded SSH host key is refused rather than trusted
- Connection parameters, client certificates, and timeouts saved per connection