Verko

Agent-first paper management.
Your papers are plain Markdown files. Your AI assistant can read them, write them, and answer questions about them.

Download · Try in browser · 中文

---

What it is

A desktop app (and a read-only web app) for organizing academic papers. Your library is a plain folder of Markdown files — no proprietary database, no lock-in. An AI agent of your choice reads and writes that library through the same files you see.

Why

• Your data stays yours. A CSV plus a folder of Markdown files. Open in Excel, VS Code, anywhere. Version-control with Git.
• AI does the busywork. Ask in natural language: "summarize my unread NLP papers", "tag the diffusion ones", "import this arXiv paper". The agent reads and writes the same files you see.
• Bring your own model. OpenAI, Claude, or Gemini — paste your API key, switch any time.
• Works online too. The web build connects directly to your S3 / R2 / B2 bucket — same UI, same agent.

Install

Desktop

Platform	Download
macOS (Apple Silicon)	Verko-X.Y.Z-arm64.dmg
macOS (Intel)	Verko-X.Y.Z.dmg
Windows	Verko-Setup-X.Y.Z.exe
Linux	Verko-X.Y.Z.AppImage / verko_X.Y.Z_amd64.deb

→ Get the latest at Releases.

Web

catvinci-studio.github.io/Verko — connect any S3-compatible bucket (AWS S3, Cloudflare R2, Backblaze B2, MinIO). Your bucket needs CORS allowed for the page origin.

Quick start

1. Launch Verko → pick Open existing folder or Create new local library (web build: Connect S3 instead).
2. Open Settings → Agent, paste an API key for OpenAI / Claude / Gemini.
3. Press ⌘K or click the Agent in the sidebar — ask anything about your library.

Your library, on disk

my-library/
  papers.csv                       ← canonical field data: title, authors, status, …
  papers/
    2017-vaswani-attention.md      ← your free-form notes (markdown only)
  attachments/
    2017-vaswani-attention.pdf
  schema.md                        ← column definitions (yours to extend)
  collections.json                 ← collection membership
  skills/                          ← optional: your agent workflow templates

papers.csv is the source of truth for every field — open it in Excel, edit a column, the app picks up the change. Each paper's .md file holds only the notes body:

## Key insight

Replace recurrence with self-attention...

## My takeaway

Read the section on positional encoding twice.

What the agent can do

Out of the box:

• Search and summarize across the whole library
• Add / update papers — direct CSV edits, or import from arXiv
• Look at PDF pages (with vision-capable models — figures, equations, tables)
• Manage collections and tags
• Take notes as you read; @-mention papers in chat to pin them to the current question

You can extend the agent with your own skills — drop a markdown file in skills/ with name + description frontmatter. The agent sees the description in its prompt and loads the body on demand.

When a conversation gets long, type /compact to compress it; the full transcript is archived. Earlier turns get summarized so the next ones stay coherent without burning context.

All file access is scoped to registered library directories — the agent literally cannot reach files outside.

Build from source

git clone https://github.com/CatVinci-Studio/Verko.git
cd Verko
npm install
npm run dev          # Electron dev mode
npm run build:web    # Static web build → dist-web/
npm run dist:mac     # or :win / :linux

Requires Node 20+. Codebase layout: CLAUDE.md.

License

MIT © CatVinci Studio