Abstract: A computer system identifies high-value information in data streams. The computer system receives a plurality of filter graph definitions. Each of the filter graph definitions includes a plurality of filter nodes arranged in a two-dimensional graph by a plurality of graph edges. Each filter node is configured to accept or reject individual packets in a data stream based on content and/or metadata information associated with the individual packet. The system prepares the filter graph for execution. In response to receiving a data stream with a plurality of packets, the system distributes the packets to inputs of each of the executable filter graph definitions. The system identifies, using the executable filter graph definitions, packets with high-value information based on parallel execution of the filter nodes included in the respective filter graph definition.