An instrument for attention.
A personal research-intelligence platform: one fully local brain that reads scientific papers end-to-end, builds a knowledge graph, and rewrites what matters into practical articles.
All inference is local
The whole pipeline runs on one host. No paper, draft or prompt ever leaves the machine — your research interests stay yours. Self-speculative decoding gives 2.1× throughput on a single GPU, with no second model to host.
Provenance on the page
Articles are openly AI-generated, and the trail sits next to the text, not in a footnote: the authoring model, the persona it wrote as, and the evaluation score it cleared. Honest, with no asterisks.
Gates, not feeds
Nothing reaches you because an algorithm thinks you will engage. Two quality gates decide: a signal score on the paper itself, and an evaluation score on the draft. Below threshold — regenerate or discard.
One brain, many voices.
The Engineer
Benchmarks, trade-offs, deployment notes. Writes like a senior colleague reviewing a paper for the team channel.
engineering_deepdiveThe Essayist
Finds the story inside the method. For results that change how you see the field, not just how you use it.
narrative_discoveryThe Skeptic
Stress-tests the claims: confounds, baselines, what the authors did not run. Healthy doubt, in writing.
critical_reviewThe Lecturer
Patient, precise explanations that survive contact with a non-specialist — without dumbing the result down.
explainer