Hundreds of papers land every day. Knowra bends that flood into a handful worth your attention — and Scout tells you in seconds whether one deserves a deep read.
How it works
Knowra fits the gaps in your day — a coffee line, a commute, the minute before a meeting.
A daily feed shaped by your fields. Swipe right to save, left to skip — every swipe sharpens what comes next.
Tap a card for a fast, structured read: the method signal, what to verify, the conditions it needs and where it breaks.
Keep what matters, with arXiv IDs at hand. Search, filter by field, and export to RIS for Zotero when you're at your desk.
Scout
An abstract is too thin to judge. A full read is too expensive to waste. Scout sits in between — a mid-level screen that surfaces what actually matters for deciding whether a paper is worth your afternoon.
Grounded in the paper itself, with sections you can trace back.
Treats attention as a nonlocal integral operator with a learned horizon δ. Tokens interact only within δ, giving a physically motivated sparsity pattern rather than a hand-designed window.
The perplexity gain at 128k tokens (Table 3) and whether the δ-ablation holds once compute is matched against sliding-window baselines.
Assumes locally smooth token interactions; strongest on code and long-form prose.
No results on retrieval-heavy tasks; custom kernel required for the reported speedups.
Why it feels different
The paper that unlocks your project is rarely the one your feed already knows about. Knowra is designed to let it through.
Five recall paths are blended and re-ranked into one feed — so you get depth in your field, and a little surprise from the next one over.
Closest in meaning to what you save
Linked through the research graph
Methods from adjacent fields
Work the field is paying attention to
What's moving right now
A middle screen that answers the real question: is this worth a deep read, and what should I check first?
Save and skip are the only settings you need. Every signal re-ranks tomorrow's feed — no boolean queries, no alert fatigue.
The same index answers your AI assistant over MCP — free paper search, no API key required.
For AI assistants
Add Knowra to Claude, Cursor, Codex or any MCP client. Your assistant searches the same library with the context of what you're stuck on right now — and brings back papers you can trace.
{
"mcpServers": {
"knowra": {
"type": "http",
"url": "https://mcp.knowra.app/mcp"
}
}
}
Set your fields once. Knowra handles the rest — every day.