FEATURES
22 shipped capabilities, from import to export. Start with the core and the five-step path, then open the full index when you need it.
01 / PRINCIPLES
InteFlow is not a study assistant that can chat about anything. It gets the certain things right first, then puts optional AI behind a key you explicitly press.
Import, parsing, grading, review and export are deterministic local computation by default; AI appears only as an explicit action, or as a fallback you have switched on.
Question banks, answers, analysis and vocabulary live under data/: one SQLite file,
plus each paper's doc.ifmd.md and images. Nothing is uploaded to the cloud by default.
data/ folder02 / WORKFLOW
No hidden sync step, and no workbench that has to be online. Every step can stop on its own — change what you need, then carry on.
Drag in PDF, DOCX, .md or .ifmd.md with a page-by-page report; scanned pages go through OCR.
Letter grid, multi-select, fill-ins and bulk paste; preview before saving, then write back to the source.
Count-up or countdown timer, question panel, keyboard control and automatic grading; multi-select and fill-in included.
Write analysis, ask AI and drag words into vocabulary in the 2-of-4 workspace; autosave in about 0.9 seconds.
Mistakes, Starred and Sessions gather your practice; export md / docx / pdf and word lists as xlsx.
03 / CORE FEATURES
Covering import, practice, review, vocabulary and export. The full set of 22 is below.
PDF, DOCX, Markdown and InteFlow-MD up to 60 MB; page-by-page reports, with suspicious pages paused for review.
Two-column layouts, question numbers, options, answer-table write-back and image extraction; no options → fill-in.
Exam-style full screen, timer, question panel and keyboard control; deterministic grading, 0 tokens.
Question, analysis, AI Tutor and vocabulary — pick two. Pages-style editor autosaves, layout is remembered.
92,006-entry offline Merriam-Webster index plus a custom dictionary; drag words in from questions, export xlsx.
Review notebooks as md / docx / pdf, whole papers as PDF / DOCX, word lists as xlsx.
.md or .ifmd.md, with page-by-page reports..ifmd.md) into My banks; add units one by one or all at once. Installers ship no copyrighted questions..eudb; a SQLite index is built on first lookup.04 / FORMAT & STACK
Every import first becomes InteFlow-MD, then goes through a deterministic state machine. No hidden second pass of cleanup, and no "looks like" guessing.
---
ifmd: 1
title: "2024 International School Final"
answers: none
---
:::q {#Q001 sec=rw no=1 difficulty=m}
@material
Despite the committee's reservations, the new policy was adopted
without amendment.
@stem
The author's attitude toward the committee is best described as
- A. undisguised contempt
- B. cautious approval
- C. outright indifference
- D. blunt hostility
! answer: B
:::# macOS / Linux
git clone https://github.com/JasonSun2009CN/InteFlow.git
cd InteFlow
./start.sh
# Windows PowerShell
.\start.ps1
# first run: create .venv, install deps, start both, open browser
# backend → http://localhost:8000
# frontend → http://localhost:5173data/
├── app.db SQLite: documents, questions,
│ sessions, analysis_pages, subjects
├── docs/<id>/
│ ├── doc.ifmd.md source of truth
│ └── assets/ extracted figures
├── dictionary/mw.db rebuildable MW index
├── exports/ notebook + document exports
└── jobs/ · tmp/ import pipeline scratch05 / QUESTIONS
Here are the answers to the things people trip over first. More detailed specs and format documentation live in the repo docs.
Ask on GitHub
No. A question's sec is a free-form label (rw, math, physics, history…), not a fixed enum. Any bank made of "question number + two or more options" can be imported; items where no options are detected are treated as fill-ins. Installers ship no SAT questions at all; the sample bank exists only in source checkouts and is excluded from packages. Import your own bank ZIP instead.
No. Import, parsing, grading, review, vocabulary and export don't depend on a network. Only AI explanations, connectivity tests, and word-exam AI grading that you turn on yourself call an external model; with no key or a failed call it falls back to local and tells you.
All under data/: app.db holds structured data, docs/<id>/doc.ifmd.md holds the original questions and answers, assets/ holds images, and exports land in exports/. To back up, copy the whole directory; to reset, delete it.
The desktop app needs no Python or Node. Running from source needs Python 3.11+ and Node.js 18+; PDF export additionally needs pango / cairo / gdk-pixbuf / harfbuzz; scanned-page OCR can optionally use tesseract, and macOS can also use Vision. When something is missing, only that feature degrades or warns.
Yes. My dictionary in Settings supports manual edits and imports from CSV / TSV / JSON / TXT, and takes priority over Merriam-Webster. For English-to-English Merriam-Webster definitions, put an Eudic .eudb into dictionary/ and the index is built automatically on first lookup. Copyrighted dictionary files are not distributed with the repo.
One review notebook per paper: a cover with the user name and score, then every question with written analysis in question order (stem, figures, your analysis). Markdown / DOCX / PDF, downloaded straight from the browser with a copy left in data/exports/.
READY WHEN YOU ARE
Do four questions online first, or download the desktop app. The main path needs no account and no API key.