A book in.
Cited claims out.
Drop a PDF and an extract.md persona. Linear jobs extract, dedupe, build concepts, then wire a graph — with a live timeline and embedding explorer.
Pipeline
LLM= Haiku / Sonnet / Opus
chunk
Split by chapter
embed
1536-d vectors
- LLM
extract
Persona claims
- LLM
dedupe
Merge near-twins
canonicalize
Unify tags
concepts
Nodes ≥ threshold
- LLM
graph
Edges + kinds
API limits
Provider headroom for distill calls.
Anthropic · Claude
Probing…
OpenRouter · embeds + fallback
Probing…
Books
0
in gallery
Live claims
0
after dedupe
Concepts
0
0 edges
Slots left
3
of 3 this week
Gallery
Finished and in-flight books
No books yet.
Take a slot →
Upload
Persona first, then PDF. Pipeline starts on success.