Press Enter to search, Shift+Enter to summarize. Paste several URLs (comma / space separated) and click Extract all to fetch & cache them — then convert to KG or vectorize for Q&A without re-fetching. Highlight any text in a summary to save it to the graph.
Cached content
Content ingested once and reusable across the KG, Q&A, Memory and summaries — no re-fetching. Add items from the Read tab; convert or vectorize them from the KG / Q&A tabs.
Ingest a corpus
Already-extracted content from the Read tab. Select items and convert them to the graph — no re-fetching. Items already in the KG are hidden.
Ask your library
Build the contextual vector library
Paste page URLs (one per line). Each page is fetched, split into sections with an LLM-written contextual summary, chunked, embedded, and indexed for hierarchical retrieval.
Vectorize from cache
Content already extracted on the Read tab. Select items to embed into the library — no re-fetching. Items already vectorized are hidden.
Ask / search your library
Focus (optional) — answer from specific documents
Leave empty to search your whole library. Select one or more documents to answer from just those.
Memory layer
What the library has learned across sessions — confirmed facts, durable answers, your preferences, known gaps, and corrections. Recall folds these into the agent and Library Q&A; answering and maintaining write new ones back.
Add a memory
Stored memories
Build a skill from context
Feed the builder the context you select — paste text, a query that matches ingested chunks, or comma-separated tags. A sub-agent pipeline understands → analyzes → authors (codeact) → evaluates → gates a new agent skill. Skills that pass the gate queue for your review; only an accepted skill joins the library.