Artifact Lab

Artifact turns gameplay into structured reasoning data for foundation models. A lightweight SDK records input sequences, reward events and decision signals, converting a session into action trajectories that carry goals, feedback and correction. A hybrid architecture drawing on R3M, Decision Transformer and JEPA/V-JEPA 2 then distils those trajectories into reasoning embeddings, low-dimensional representations of human intent, planning and adaptation. Outputs run in two directions: personalised in-game intelligence for publishers, and reasoning-grade training data for AI labs. The positioning is a gameplay data layer, distinct from GameFi projects that issue on-chain assets.

What separates this data from text and image corpora is the trade-off it captures: players test strategies under resource constraints and revise them, leaving an aligned decision sequence. The company began with in-game user profiles and advertising monetisation and now describes itself as powering models rather than building them. Its 2026 product pages set out three stages, capture, distillation and activation, with separate onboarding for studios and AI labs, and stress that trajectory order and reward signals stay reusable downstream.