Turn your PDFs into knowledge you can search and apply.

PDFs turn into notes you can edit with the tables, graphs and the maths preserved. AI can help you to find what you need quicker and even build prototypes.

Get started

PDFs in , knowledge out.

From PDF to text

Drop the paper in, turn the scan into words, read it, then ask your AI about it.

The Upload files dialog in Oyren, with a drop area, a field to paste a link to a PDF, and Cancel and Done buttons

Ask, and get the passage back.

Pin the papers, ask in plain words, and read the line the answer came from.

  • Find the paper.

    Searches the web for papers and PDFs, and saves the ones you keep.

    Find Papers
    step size bounds for mirror descent
    Each search uses ~15 credits.PAPERS
    Mirror descent under Bregman divergences
    arxiv.orgSep 4, 2026OpenSave PDF
    A survey of proximal splitting methods
    openreview.netJul 22, 2026OpenSave PDF
    FROM THE WEB
    Lecture 9: Bregman geometrystanford.edu
    Saved papers land in your files, ready to convert.

  • Ask what you pinned.

    Answers from the papers on the pin rail, quoting what it read.

    Convex optimisationSonnet
    Pinned context (2)mirror-descent.pdfbregman-scan.md
    @proximal-survey.pdf does the survey use the same step size as section 3.2, and does the entropic case change it?
    Two answers.The survey uses the same 1/L, and calls it the standard choice for an L-smooth objective.“For the entropic regulariser the bound holds unchanged provided ψ is 1-strongly convex on the simplex.”bregman-scan.md
    Context24,180 / 200,000 tokens (12.1%)
    Charged on real token use · 10,000 credits = $1

  • Let it find the passage.

    Ranks your files, reads the finalists, and hands back one quote.

    One of these papers proves the bound holds on the simplex. I cannot remember which.
    Candidates (top 8) · keywords: simplex, strongly convex, bound
    mirror-descent.pdfscore 0.87
    bregman-scan.mdscore 0.72
    Deep-read sub-agent results
    bregman-scan.mdyes (92%)
    mirror-descent.pdfuncertain (41%)
    proximal-survey.pdfno (8%)
    “…provided ψ is 1-strongly convex on the simplex, the same bound holds.”bregman-scan.md, attached to this conversation
    Is this the right file?YesNo
    Nothing runs until you approve the search.

When prose is not precise enough.

State the lemma in Lean 4 and let the kernel judge it. The Lean 4 Power App opens real VS Code in the browser with Mathlib and its prebuilt cache already there, so import Mathlib.Tactic works on the first line. No elan, no lake, no multi-gigabyte download.

Claude Code or opencode sits one pane over and drafts tactics. Nothing it writes counts until Lean has checked it.

See the Lean 4 workspace
POWER APP · LEAN 4your own container · Medium or larger
Tutorial.leanVS Code
import Mathlib.Tactic-- `ring` closes algebraic identities over ℤexample (a b : ℤ) : (a + b) ^ 2 = a ^ 2 + 2 * a * b + b ^ 2 := by ring-- an open exercise, for you to finishtheorem my_add_succ (n : Nat) : n + 1 = n.succ := sorry
Lean 4 · mathlib · elaborated in 240ms
LEAN INFOVIEWTutorial.lean:5:47
Goals accomplished
ALL MESSAGES (1)
Tutorial.lean:8:41declaration uses ‘sorry’
Goals, hypotheses and errors update as you type, the way Lean users expect.
Open a workspace, open the Lean 4 Power App, state your theorem. That is the whole setup.

Bring one paper and see what it reads like.

The workspace is free to keep. You pay only for the runs you ask for, and each one shows its price before it starts.

Get started