The ability to map cellular interactomes and organelle proteomes is foundational to a molecular understanding of cell biology. Proximity labeling (PL) provides a powerful strategy for such mapping, but existing enzymes and photocatalysts are limited by their reliance on biotin, spatial resolution, and/or in vivo compatibility. Here we report FlexID, an engineered promiscuous ligase that rapidly attaches diverse small-molecule probes to nearby cellular proteins through direct contact. We engineered FlexID using both sequence- and structure-trained computational models, combining their strengths for catalytic enhancement and structural stabilization. Biophysical analysis reveals that conformational changes in FlexID improve its ability to recognize diverse target proteins while preventing premature release of reactive intermediate. We demonstrate FlexID’s versatility through in vivo proximity labeling, comprehensive organelle proteome mapping, and high-throughput screening of molecular glues using fluorescence-based readout. Our work demonstrates that computational methods can be harnessed to create mechanistically distinct PL enzymes, and provides a flexible, high-resolution tool for mapping protein interactions and proteomes in living cells.