Intelligence
Tools that help SQLly explain your database in plain language without pretending it knows more than it does.
A schema can tell you what a column is called; it cannot always tell you what the column means. SQLly's intelligence work is about closing that gap carefully. The live value-aware completion features are under IntelliSense › Data awareness; this section covers the layers built on a language model.
Everything is grounded in your catalog
No feature here asks a model to guess at your database. SQLly keeps a searchable catalog of your schema objects and, for each question, assembles a bounded slice of it: the objects ranked as relevant, trimmed to a token budget, with the list of what was included carried back alongside the answer. So a response comes with its own citation, you can see which tables and columns it was built from.
Where a deterministic check can answer instead, it goes first and the model gets its findings. SQL review, for instance, runs the parser-backed validator, then asks the model to explain and address each real finding, which is why the narrative talks about your query rather than about SQL in general.
Which model runs it
| Where it runs | Options |
|---|---|
| On device | Ollama on any platform, or the MLX runtime on Apple Silicon. Nothing leaves the machine. |
| Your own endpoint | Any OpenAI-compatible or Anthropic-compatible base URL - a self-hosted gateway, a model server on your LAN. SQLly labels a profile pointing at localhost as local. |
| Cloud, if you ask | OpenRouter, OpenAI, Anthropic, Gemini, or DeepSeek. Each is a named provider profile with its own model, context window, reasoning level, and optional proxy - and its key stored as a secret reference, not in the config file. |
local or cloud.Trying it before trusting it
The AI playground lets you point each provider profile at a prompt and compare what comes back, same question, same catalog context, different model, with the token limit under your control. It is the honest way to find out whether a small local model is good enough for the job before you wire it into your daily work.
In this section
- On-device query summarization in progress
- Get a plain-language preview of a query from a model you control, before you have to parse every clause yourself.
- Auto-magic string documentation planned
- Trace an unexplained value through DDL and leave behind a useful note for the next person, possibly future you.
- Native grep-ai search planned
- Search across your schema and data by intent, not only by the exact SQL words you happen to remember.
- Decision model
- An optional fast cloud model that answers tiny pick-one questions so several features can offer a smart first guess.
- MCP server for coding agents
- Let Claude Code, Cursor, VS Code and Windsurf use your saved connections read-only over MCP, with no passwords shared.