AIWright

AIWright is a dialogue focused AI toolset for interactive narratives. It clusters large dialogue graphs, generates grounded lines by masked infilling inside Lua scripts, runs mobile friendly playtests, and analyzes why players chose one option over another.

AIWright is a research-oriented tool for teams that build interactive dialogue: narrative designers, game writers, and researchers who need more than a chat box. It treats a conversation as a graph, not a single prompt. The system analyzes large dialogue trees, clusters nodes by any attribute you care about, and surfaces structure that would be tedious to trace by hand. That same structure feeds generation, so new lines are produced with knowledge of where they sit in the tree rather than in isolation. The generation approach is unusual. Selected portions of a dialogue tree are converted into a Lua script, and the model fills in masked sections of that code. Writers keep the branching logic, conditions, and variable checks in a language the game engine already understands, while the model handles the natural language inside the mask. The homepage demonstrates this with Disco Elysium style passive skill checks and variable gated options, which shows the tool is aimed at systems heavy RPG dialogue rather than linear scripts. Playtesting closes the loop. A mobile friendly web interface lets players move through grounded conversations, and the same setup can be used to compare how people respond to different model authored variants. After a session, an analysis suite helps compare models and dig into why a player picked one option over another, which turns subjective writing judgment into something closer to evidence. The team has published research on player perceptions of GPT-4 generated dialogue in Disco Elysium: The Final Cut, so the methods are documented rather than hidden. AIWright fits best alongside narrative design tools and general purpose writing assistants. It is not a prose generator for novels; it is closer to a lab bench for branching dialogue. Teams evaluating it should be comfortable with Lua style scripting and graph concepts.