What problem does it solve? Raw tabletop RPG session transcripts arrive as fragmented CSV files with unresolved speaker labels, mixed in-character and out-of-character chatter, and no structure — making it hard to turn hours of audio into usable campaign wiki content. ## Core Features & Use Cases - Speaker Resolution: Maps generic "Speaker N" and "Unknown" labels to known players using conversational context, process of elimination, and temporal clustering, then retrains voice profiles for future sessions. - Canon Extraction & Recap: Processes each transcript part incrementally, classifying lines as IC/OOC/META, merging fragmented utterances, and producing tagged extracts ([CANON], [LORE], [COMBAT], [ITEM], [SIGNAL], [NPC], [RULING]) with line-level citations. - Wiki Integration: Folds recaps and extracts into an Obsidian campaign wiki, updating session notes, entity pages, combat analytics, and the current world-state file. - Use Case: After recording a 4-hour D&D session, run the ingest to resolve who said what, produce a scene-by-scene recap with structured per-PC combat tables, and update the campaign wiki — all checkpointed so work survives context limits. ## Quick Start Ask the agent to process the session 12 transcript and it will resolve speakers, extract canon, and update the wiki.
