Multi-Agent Storyworld Simulator
Built a resumable multi-agent narrative engine with per-character memory and SQLite checkpointing.
Project Overview
Built a stateful multi-agent narrative simulation engine that coordinates autonomous character agents, environment agents, and story workflows. The system uses memory, state-graph orchestration, and SQLite checkpointing to support resumable, goal-directed episodic stories.
Project Description
Narrative generation systems often lose continuity, character memory, and resumability across longer story sessions. This simulator focuses on stateful story workflows where agents can remember events and continue from saved checkpoints.
- Orchestrated specialized agents across initialization, environment, and character layers.
- Added short- and long-term character memory so agents can reference prior events.
- Technical approach: built state-graph workflows with autonomous character agents, environment orchestration, Pydantic validation, and SQLite checkpointing.
- Outcome: created resumable, goal-directed episodic workflows where characters can reference prior events and maintain state across sessions.
Technologies Used
PythonPydanticLangGraphSQLite
Project Links
Project Gallery
