Monet is an extensible database kernel under development since 1993 at CWI. It is amongst the first full-fledged implementations designed from the premisses that main-memory considerations should drive the architecture. That is, all but the largest databases are memory resident. This system forms the basis of a large national project to develop technology to index multi-media web sources.

In this talk I will give a short overview of its architecture and modelling technique to encapsulate proprietary and open-source multi-media feature extractions. The key innovation is to regard multi-media indexing as an enlarged parsing problem, which can convientely be described with a context-free grammar. A small toolkit based on traditional compiler-compiler technology then forms the bridge to store the index information in the Monet database, while the grammar itself provides the namespace to query it.