sci-fork

In biomedical research, questions, hypotheses, literature evidence, lab results, and findings tend to scatter across notes, PDFs, and chat threads, with no connected, inspectable view of how they relate. SciFork plugs into DeepSeek Harness as a local, Git-native research graph: your project is simply ordinary Markdown and JSON files in a local Git repository, the graph is a rebuildable view of those files rather than a separate database, and nothing is ever uploaded or cloud-synced.

Core capabilities include importing PubMed evidence by PMID or DOI, keeping untested hypotheses strictly separate from team results and human-reviewed findings, switching between a Main overview and a focused Evidence view for any entity, running a bounded one-step literature-driven expansion (capped at five low-confidence branches) via Research & Expand, and writing a local Git checkpoint after each successful change that touches only SciFork-managed files.

It is built for individual researchers or small teams doing literature-driven biomedical inquiry or hypothesis-driven research within DSH, especially those who need structured, traceable progress tracking while keeping all data on-device and free of external databases or cloud dependencies.