[fuseraft]

An agent had no way to tell which SQL dialect a database speaks, and one query could return about a million tokens, more than a model's whole context.

• Providers. list_databases returns [{"name":"prod","provider":"Postgres"}, …], and its description tells the agent to write each database's SQL in that provider's dialect. Before, the agent learned TOP vs LIMIT vs FETCH FIRST from a syntax error.
• Rows as arrays. Each row is an array of values in the order of columns, so column names aren't repeated on every row. /api/query returns the same shape.
• Smaller defaults. App:MaxRows goes from 1000 to 100, and App:MaxResultBytes from 4000000 to 100000 (about 25,000 tokens). The new App:MaxTables (default 1000) limits list_tables, which MaxRows used to, so the table list isn't cut to 100.
• Truncation notes. A truncated result has a note saying which limit stopped it and how to narrow or page the query, with paging in the database's own syntax:

{ "columns": ["n","sq"], "rows": [[1,1], …, [100,10000]], "truncated": true,
  "note": "Only the first 100 rows are shown (App:MaxRows). Narrow the query with WHERE, summarize it with GROUP BY, or page through it with ORDER BY ... LIMIT 100 OFFSET 100." }

Tested locally: the full suite against all four databases (319 tests), and a 500-row query over MCP against Postgres, which now returns 100 rows in 1,172 characters with the note.