arXiv Research Digest

100% local — Ollama at http://localhost:11434 · LLM zai-org/GLM-5.3-Flash · embeddings BAAI/bge-large-en-v1.5

✅ Ollama reachable. Pulled models: zai-org/GLM-5.3-Flash, BAAI/bge-large-en-v1.5

Days back

Top N

Max fetch

Embed→LLM budget K

deep digest top-N

Ranking mode Hybrid — embed all → keep top-K by similarity → LLM scores those → rank (recommended) Embeddings only — pure similarity ranking, no LLM (fastest; no significance judgment) Straight LLM — LLM scores every fetched paper (slowest on local models)

Field

Subfield (Ctrl/Cmd-click = several; none = whole field)

Topic / research area (used as the embedding query & LLM focus; advanced syntax like ti:"world model" AND cat:cs.RO also works) ✨ formalize this into precise arXiv terms with a small local LLM (great for vague phrasing; auto-skips when you already wrote arXiv syntax)

Custom arXiv categories (optional, comma-separated)

also hard-filter arXiv by topic text

hide papers I've already seen ·

⏳ Fetching → embedding → shortlisting → LLM digesting… on a local model this typically takes 2–6 minutes.