hivein for Postgres
Ask your Postgres database questions in plain language. Every table becomes a tool, and joins come from your foreign keys.
Host, port, database, username and password, with SSL on or off and an optional CA, client certificate and key.
- Every table becomes a tool
- Counts and totals without paging
- Joins from your foreign keys
- Writes only when you turn them on
- No raw SQL, bounded reads
Your Postgres, as actions the copilot can take.
Every table becomes a tool
Each table gets a select tool with filters, ranges, IN lists, null checks, substring search, ordering and a column list, so a question reads the rows it needs and nothing more.
Counts and totals without paging
An aggregate tool per table does count, sum, average, minimum and maximum with group-by, so a question about how many is one query, not a thousand rows read and counted.
Joins from your foreign keys
Foreign keys are read when the source connects and become joined reads, and pure join tables become a list of related ids on the tables they link, so a question can cross tables in one call.
Writes only when you turn them on
Insert, update and delete tools exist only if writes are enabled for the source. Update and delete work by primary key, and every delete asks first.
No raw SQL, bounded reads
Queries are built from typed specifications, never from text the model wrote. Reads return 100 rows by default and 1,000 at most, and a multi-step write runs as one transaction.
Every Postgres source starts with these.
Written in plain language, not configuration — change a sentence and the behaviour changes with it.
Counts come from aggregates
Totals and counts are asked of the aggregate tool, never worked out by paging through rows.
Writes report what happened
A write reports the id, the affected rows and the row as written, and a related-ids list replaces the whole set of links rather than adding to it.
It reads your Postgres data model, not a generic one.
The schema is read first, so a question that spans two of these is one question, not two lookups.
“How many orders did we take each week this quarter?”
“Which customers have not ordered in the last six months?”
“What is the average refund per product category?”
“Show the ten largest invoices that are still unpaid”
Writes to Postgres wait for you.
The same three guardrails hold on every template. They are attached to the action itself, so they hold on every path. Watch the replay →
- 01
The plan is held
A plan that writes to a real system stops before it starts. You see the goal, the risk and every step, then confirm or reject.
- 02
Each write asks first
A tool marked ask-first shows the exact call with its values. Confirm it, edit the values, or reject it. Nothing runs until you decide.
- 03
A rejection is an instruction
Pick a reason and add a line. The copilot re-plans with what you told it instead of guessing.
Postgres questions.
Can I limit which tables it sees?
Yes. A database source can be limited to chosen schemas and tables, and tools are generated only for what is in scope.
Does it work with a managed database?
Yes. SSL can be switched on without certificates for managed services such as RDS or Supabase, or given your own CA, certificate and key so the server's identity is verified.
Will it put load on the database?
Each source keeps at most five connections, and every read carries a row limit. Statements time out after 30 seconds.
More templates like this one.
MySQL
Open sourceAsk your MySQL database questions in plain language. Every table becomes a tool, and joins come from your foreign keys.
Tools from your schemaOpen template → — MySQLConnect Postgres. Get a copilot this afternoon.
No card, no agent configuration, no prompt to write. Sign in and approve the actions it may take.