A sovereign research engine you can talk to
Three ways to use it — ask the shared library, turn your own documents into a living map you can question, and package a reusable, citation-grounded skill your agents can call.
Retrieval-augmented generation couples a generative model with an external memory it can query at inference time: it retrieves relevant passages from a searchable index, then feeds them with your prompt to the generator, producing an answer explicitly grounded in the retrieved evidence 12. By anchoring generation in real documents, it dramatically lowers the chance the model fabricates plausible-but-false statements 3, and because the retrieved sources are retained alongside the answer, every claim can be traced back to a verifiable document 4.
Conflict-based approaches assess road safety from near-miss interactions — traffic conflicts — instead of waiting for years of crash records, which makes them proactive and usable in near-real time 12. Conflicts are captured from video or extracted from microsimulation, then linked to crash frequency using extreme-value theory 3. Unlike traditional crash-frequency models — reactive and dependent on long data-accumulation periods — conflict-based methods can even evaluate designs before they are built, and the literature increasingly fuses the two to combine their strengths 4.
Every claim carries an inline citation you can click and verify. No hand-wavy summaries — it tells you which source said what, and where it inferred across them.
hybrid semantic + keyword recall · cross-encoder reranking · scope-aware synthesis
Upload your whole library or sync Zotero and question it like a colleague. Not one PDF at a time — thousands of documents, indexed and searchable together. NotebookLM stops at a handful; a single chat can't hold your corpus. This can.
per-user isolated index · scales with your plan · Zotero & direct upload
Your documents are encrypted and only accessible by you. Every account is fully isolated — no one else can read your files, and they're never resold or handed to a third party. Your library stays private, exactly as it should be.
encrypted · per-user isolation · only accessible by you
Built for anyone who works with dense documents
A sovereign pipeline that builds a citation-aware knowledge graph — then reasons across it to answer your question.
Autonomously surfaces relevant papers and captures how they cite one another.
Parses each PDF into overlapping, section-aware passages — then deletes the file.
Embeds every passage and indexes it two ways — for meaning and for exact terms.
Links papers by citation and similarity, distilling their findings into a reasoning graph.
Reads your question and picks the right retrieval depth and strategy.
Searches by meaning and by keyword at once, then fuses the two rankings.
Walks the citation graph for related work — scope-checked so different contexts never blur.
A cross-encoder re-scores every candidate so only the strongest evidence survives.
Writes a grounded answer with inline citations — stating what each paper claims versus what it infers across them.
Start free — no card. Upgrade when your research outgrows it.
Try it out, no card required
For regular research sessions
For serious, daily research work
For power users & teams
Sequential questions use 2x your daily budget; Sequential + Section mode uses 4x; Skill Mode uses 3x. Prices in USD. Cancel anytime. Secure checkout via Stripe — Apple Pay, Google Pay & cards.
No signup, no card. 3 free questions to see it work — then higher usage limits with a free account.