Scholardo is the research workspace for macOS. Your papers, highlights, notes and drafts stay on your own disk, and the agent works inside them — built in, or your own Claude Code.
Two researchers, one workspace. Most tools treat the agent as a caller — you ask, it answers, then you go back to working alone. Scholardo treats it as a resident: same documents, same annotations, same live context. It knows which file you have open, and which page you are on.
$ claude > what is this page arguing? ⏺ scholardo_get_active_context ← tab: sources · preview: Kahneman_2011.pdf · page 47 ⏺ scholardo_search_annotations ← 3 highlights on this page, 1 note from March Page 47 sets up the anchoring experiment. You highlighted the sample-size caveat back in March and flagged it as the weak point — the argument here still rests on it.
Use the agent built in, or bring your own CLI
The built-in agent needs no terminal and no setup beyond choosing a model, and has the same access to your library as any CLI here. Claude Code and Codex are their vendors' official CLIs and need their own subscription or a compatible third-party model; Scholardo is not affiliated with them. Open-source CLIs like opencode and pi run on any protocol-compatible model provider.
Each step leaves something the next one can use — nothing is exported, re-imported, or re-explained.
Search the major research databases at once, drop duplicates, and put the most relevant first — then save a free, legal full-text PDF in one click. Your agent can search too, over MCP: Semantic Scholar, arXiv, or any server you connect.
Browse the web without leaving the app and save any page — or a PDF — straight into your library. Your agent can read whatever you are looking at, too.
Follow journals, blogs or news by topic and get a short AI summary of what is new dropped into your library on a schedule.
Paste a link or an ID and the journal, year, authors and abstract fill themselves in from several sources — blanks only, never overwriting what you typed. Import a batch from the tools you already use and it is split into one reference each, named consistently.
PDFs, e-books, Office files, web pages, images, audio and video — all in one place. Search within a file and pick up where you left off.
Highlight and annotate PDFs, e-books and web pages with one set of tools, and jump between marks across chapters from a sidebar.
Already highlight on the web with Hypothesis? Pull those notes in automatically, sorted by project — including ones you made before you added the document.
Annotations are keyed to the content itself, not the filename. Rename it, move it, reorganize your whole library — the marks stay attached.
Keep notes that link to each other so you can follow a thread of ideas — and see the whole shape of it as a graph.
Find things by what they mean, not just the exact words — so searching for one idea surfaces the related passages too, across everything in your library.
Every note remembers the session it grew out of. Six months later you can still trace an idea back to the conversation that produced it.
As it works, the agent writes what it learns back into your library — your conventions, your open questions, the thread you were pulling last week.
Write a paper in chapters that stay organized on their own, so a long draft never turns into a pile of files.
Every document carries a writing brief — a statement of intent you and your CLI agent both maintain. Select any text and the AI rewrites, proofreads or drafts against it, so a polish never quietly changes your position.
Choose from 100+ citation styles — APA, MLA, Chicago and more. Scholardo builds the bibliography, ready to paste straight into your paper.
Scholardo checks that every document has a reference and every reference has its file, and links each reference to the right PDF.
Turn your notes into a deck and pick from 36 themes. Present full-screen and export to PDF — no separate slideshow app.
Mermaid diagrams, KaTeX maths, charts, mind maps, pseudocode — even chemical structures and music notation. Written as plain text in your notes, rendered in the deck and in every export.
Figures, tables and equations number themselves and renumber when you reorder a section. Citations inside a deck resolve against the same reference library the manuscript uses.
Export to PDF, Word or LaTeX through pandoc with your references and formatting already in place — the bibliography is not a separate step at the end.
18 rules, all on your own machine — no network, no model. They run at export: a document that fails can still go, but the waiver is recorded.
A [@key] that matches nothing in your library; a key that exists but not in the .bib the document declares, so the citation would never render; the same key defined twice; an entry sitting in the bibliography that the prose never cites.
If a citation says page 47, the local copy is measured to see whether it has a page 47. A cloud tool cannot do this one — it does not hold your file.
A figure, a year or a named attribution with no citation; a quotation with no locator; a borrowed exhibit that does not say where from; and a source you cited but never read.
When it checked, against which rule set, the hash of the text, how many errors and warnings — and which checks did not run. ICMJE holds the author responsible for verifying AI-assisted content; this is the written form of having done so.
~/.scholardo/sources/Collect → your PDFs and pages, untouched~/.scholardo/annotations/Read → sidecar JSON, keyed by content — survives rename and move~/.scholardo/notes/Think → plain Markdown, each one remembering its source session~/.scholardo/refs/Write → standard citation formatsEach project is backed up on a schedule and kept offline, and an interrupted edit is replayed the next time you open it.
| Capability | Scholardo | Zotero | Mendeley | Obsidian | EndNote |
|---|---|---|---|---|---|
| Native agent tools (MCP) | |||||
| Dense semantic search | |||||
| Lexical + semantic fusion | |||||
| Unified PDF/EPUB/web annotations | |||||
| In-app browser capture | |||||
| RSS feeds + AI digest | |||||
| Citation rendering (CSL 100+) | |||||
| Agent sees your live reading position | |||||
| Cross-session project memory | |||||
| Bring-your-own model (multi-provider) | |||||
| Local-first · Zero telemetry | |||||
| Session branching & merge | |||||
| Menu-bar todo + AI summary | |||||
| Hypothesis annotation sync | |||||
| Presentation / slide output |
Bring your own agent — no credits, no markup: you pay for the workbench, never for the AI running on it. Every price below is a temporary early-bird price, not a permanent one.
All prices in USD · Taxes may apply at checkout
Scholardo is priced against what it does today, and every milestone on the roadmap makes it do more — so the price goes up as those land. Both plans hold their founding price through September 30, 2026.
Your research data, no. It stays in ~/.scholardo/ on your machine — no telemetry, no usage data, no cloud sync. The only calls we make on our own are the license check and, on this site, a version lookup; features you trigger yourself (an AI provider, an RSS feed, opening a page) reach out to whatever you pointed them at. The privacy policy lists all of it.
No. Core features — search, annotations, citations, browser capture — work locally without any key. A key is only needed for AI-powered features like selection rewriting or feed digests. You can also route different kinds of task to different providers and models, or point everything at a local Ollama model.
Before you export, 18 local rules check the chain from each claim back to its source: a [@key] that resolves to nothing, a page number past the end of the file you cited, a quotation or a figure with no source, and a reference you cited but never opened. Running them needs no network and no model — nothing leaves your machine. A document that fails can still be exported, and the waiver is recorded along with which checks did not run.
Zotero is built for one researcher. Scholardo is built for two: you and your agent share the same library, the same annotations, and live reading context — natively, not through a plugin.
Those let an agent read your references. Scholardo lets one work inside your whole research: not just papers and metadata, but your highlights, your notes, your drafts, and whatever you have open right now. The difference is not whether an agent can reach the library — it is how much of your work the library holds.
Agent CLIs live in a terminal, and this is where that terminal is — we follow the agents rather than the platform count. Being native is also what lets Finder tags sync both ways and lets the app reach your files without a sync layer in between. There is no iOS, Windows, Linux or web version, and none is planned.
Almost entirely. Local search, annotations, citations, and terminal sessions all work offline. Exceptions: fetching web content, cloud-model embedding / AI calls if configured, and the annual license refresh. If you switch to local open-source models via Ollama, embedding and AI generation run on your own machine — so semantic search, selection rewriting, and more also work fully offline, unlocking even more without a connection.
Personal licenses come in two forms: Annual ($49.99/yr, cancel anytime) and Lifetime ($199.99, pay once). Both are Founding Researcher prices, held through September 30, 2026 and rising as the roadmap lands — a subscription started during the program keeps its rate for as long as it runs without a gap, and whatever you have paid in subscription comes off the lifetime price if you switch. Download the app and unlock it with a license key. Team pricing is available on request.
From the merchant of record that took the payment, not from us. Buy through Stripe and the receipt and invoice PDFs arrive attached to the confirmation email; buy through Creem — the route for mainland China — and you download them from the order page linked in its order email. Either way they show the amount, the tax and the order number. One caveat that only affects mainland China: it is a commercial invoice from a non-Chinese entity rather than a Chinese VAT invoice, and checkout has no field for a Chinese company name or tax number. Get in touch if your finance department needs supporting paperwork.
Up to 3. A fourth activation is refused until a seat is freed — open Scholardo → About on the Mac you want to release and choose “Deactivate this device”. The seat frees up immediately, and that Mac can be activated again later.
No. An agent is built into the app — choose a model and start; it can do everything a CLI can do with your library. If you would rather use the Claude Code or Codex you already pay for, that works too, from a shell session.
Free 7-day trial · no payment required · all features
Tell us your team size and use case — we'll get back to you by email with pricing and onboarding.
Got it — we'll email you shortly.