Magma

Companies building AI agents generate large amounts of interaction data every day: traces, corrections, and completed multi-step workflows. This data reflects how real work actually happens inside real products, which AI labs want and pay for. Today, according to Magma, this data is simply thrown away rather than captured or monetized. Companies have no simple mechanism to package, scope, and license this data to interested buyers.

Magma onboards companies building AI agents and sets up a read-only, encrypted, scope-approved connection to their data sources (shown in the example as a production Postgres database). It packages interaction data such as traces, correction bundles, and multi-step workflow sessions into licensable datasets for AI labs and researchers. The company's dashboard tracks connected sources, trace counts, packaged datasets, and monthly earnings, and shows licensing activity as bundles are licensed or placed in review. Customers earn a revenue share as their packaged data licenses, with terms set out in a data agreement signed during onboarding; the process is described as a two-minute form plus a twenty-minute onboarding call.