/init

To build a research wiki from existing PDFs and notes, you can bootstrap a structured wiki by normalizing raw input paths into prepared artifacts and scaffolding the wiki structure for parallel paper ingestion.

Parallel paper ingestion works by scaffolding the wiki structure first, then delegating one-paper-per-subagent work to isolated worktrees, and finally merging results safely with deduplication and rebuild steps.

Yes, you can use optional planner-guided external discovery to find related papers. The system ranks and trims candidate papers to a final target set of typically 8–10 before fetching and ingestion.

If some papers fail during wiki ingestion, the resilient recovery system checkpoints each stage, continues through partial failures, and refreshes visualization artifacts without failing the entire initialization run.

You need raw input paths in directories like raw/papers/, raw/notes/, and raw/web/. The system handles deterministic source preparation by normalizing your PDFs into artifacts and writing recovery-aware manifests for traceable re-runs.

To ensure traceability when re-running initialization, the system writes recovery-aware source manifests and checkpoint manifests at each stage, allowing you to track prepared artifacts and resume from specific points.