Event Deep Research: an open-source project that builds chronologies (github.com)

Event Deep Research is an open-source agent system that automatically researches historical figures and outputs clean, structured JSON timelines. Given a query like {"person_to_research":"Albert Einstein"} the pipeline crawls sources, detects biographical events, and returns normalized events (name, description, date object with year/note, location, ISO-like id). The project uses a Supervisor Agent to orchestrate Research, Think and Finish tools, a URL Crawler (Firecrawl) to extract web content, and a Merge Agent to combine and deduplicate events — with optional image attachments and real-time visualization via LangGraph Studio (open at http://localhost:2024). Technically it’s built for Python 3.12+, uses the uv package manager, and supports OpenAI, Anthropic, Google, or local models (Ollama). Configurable model roles let you pick separate models for structured output, tool calls, and chunk-level event detection; there are controls for chunk/overlap sizes, max tokens, retries, iterations and max chunks. This makes it useful for dataset curation, temporal knowledge extraction, and digital humanities work while preserving provenance through merged sources. Caveats: quality depends on upstream crawlers and LLMs (API keys required for production), so users should review sourcing and biases. The repo is MIT-licensed and encourages contributions.