What does plainDB’s on-device AI require?
plainDB's on-device AI assistant is a Mac feature. It requires a Mac eligible for Apple Intelligence, Apple Intelligence enabled in System Settings, a downloaded and available model, and the assistant enabled in plainDB settings.
Why might the assistant be unavailable?
The app checks the system model's availability. It distinguishes an ineligible device, Apple Intelligence being switched off, and a model that is not ready yet. Installing a compatible macOS version alone does not guarantee that the hardware or model meets those requirements. The assistant offers a route to System Settings and a way to check availability again.
The implementation uses Apple's Foundation Models framework. Windows has a local assistant with the same request and SQL-suggestion workflow. Enable it in plainDB Settings, enter the installed model name and provide an OpenAI-compatible loopback endpoint. The default is http://127.0.0.1:11434/v1/chat/completions. A model server is not bundled; remote endpoints, proxies and redirects are rejected. You can still use the database browsing and SQL workflow without AI.
What information does it use?
The assistant receives a compact schema summary and context relevant to the requested action, such as current SQL, the last database error or EXPLAIN output. You can optionally include the first 20 rows of the current result. A limited context window means it may not see the entire schema or conversation.
Can I run its suggestions?
The interface can insert generated SQL into a query tab and offers execution controls. Review the statement, database, role and affected records first. A generated query may be wrong even when it is valid SQL. Use restricted credentials and test changes on disposable data where appropriate.
The current SQL context is geared to PostgreSQL and SQLite; do not assume equal AI dialect support for every connected engine. See editing safeguards and the platform feature matrix.