Lighthouse

A document workspace for research work. The researcher drives search, reading, citation lineage, and gap-network work.

The AI stays alongside that flow. It might add a short interpretation to a search result card, unpack a figure or table inside the open PDF into prose, or thread a citation lineage into a continuous flow. The tools are started by the user; the AI surfaces a short reply inside the screen that helps the next move.

What it helps with

• Search — Doesn’t stop at keyword hits. Result cards summarize the paper landscape.
• PDF reading — Figure/table interpretation, failure recovery, and per-document AI reaction rules.
• Citation lineage — Opens references and citations together to trace flow and lineage.
• Gap network — Detects gaps between paper clusters surfaced by search.

What each feature promises

Each feature has the behavior the AI guarantees to the user spelled out as named items. The item, the situation in which it triggers, and the larger activity it sits inside narrow from top down. What to guarantee is decided by a human.

Rules that have to hold across multiple features — like “no vague phrasing” or response-length caps — are pulled out as separate items and applied consistently across all relevant features.

Each item is verified along two strands: a deterministic check and a meaning check. The release verdict shows up as separate items, not a single pass/fail line.

This operating shape is the sample for How I work.

Where it stands

Closed-beta.