Indelis turns the sources you've read and verified into original work — where every claim traces back to evidence you can defend.
Saving was never making. This is the layer where the reading finally becomes work.
Free to self-host · your data and models, local or cloud.
A year of reading decays to roughly a third by the next day (Murre & Dros, 2015) — and to almost nothing you shipped. The source that disagreed with the claim under your conclusion, you averaged into false consensus. For someone who must stand behind their work, that isn't a lost note — it's a defective deliverable.
"That's the knowledge graveyard — not notes nobody kept, but work nobody made. And it isn't your shame; it's your raw material."
Unverified notes are weight, not foundation.
Connection happens between things you've internalised, not files.
Synthesis needs the focus the tools keep stealing.
23:15 to refocus after one interruption (Mark, CHI 2008).
Every claim traces to a source. Ungrounded sentences get flagged, not dressed up as fact.
Contradictions surface as angles, not alarms. This is calibrated doubt — the thing the analyst actually pays for — not a confident-sounding answer.
Weaver proposes a non-obvious bridge between two things you already hold and verified — with auditable reasons attached.
Each proposal carries its own sub-scores and the cognitive-science methods it rests on. The machine proposes; you accept. That's the difference from "just ask ChatGPT".
A behaviour without a methodology citation fails CI — it cannot ship. 32 of 41 methods are wired and load-bearing on live paths today; 5 in progress, 4 planned.
Ebbinghaus (1885) / Murre & Dros (2015) on forgetting · Koestler, The Act of Creation (1964) · Gentner (1983) structure-mapping · Mark, Gudith & Klocke (CHI 2008) on the cost of interruption.
Grounded synthesis: work you shipped whose claims trace to notes you verified.
Time is in the denominator — it counts against the score, never for it. No streaks, no feed, no session counters: nothing fights for the attention synthesis needs.
Your data doesn't leave, doesn't train anyone's model, isn't sold. No machine-authored synthesis with your name on it — the exact failure this whole page is about, refused by design. Full Markdown export, any time.
We speak in the present tense only about what works on a wired path. The labels on this page match the labels on every screen above.
Built by someone who needed it — I'm the person this is for. Follow the build on GitHub.
For researchers, analysts, and anyone whose name goes on the work. We reply within a couple of days.