# Flowing

> Let the papers you read shape what comes next. Flowing is a desktop academic writing environment that brings manuscript context and inspectable passages from your own PDF library into the writing process.

Best suited to researchers writing literature reviews, papers, or thesis chapters who already have relevant PDFs and want to recall source passages, check claims, and move the argument forward. It is not a general chatbot, reference manager, or paper discovery engine.

Chinese summary: https://flowing.works/llms-zh.txt
Full documentation: https://flowing.works/llms-full.txt

## Three core capabilities

### Snippet Card: find the passage you remember

Recall identifies keywords in the draft and surfaces relevant passages from selected local PDF libraries as Snippet Cards, with excerpts and source-page access. The researcher can return directly to the original PDF, rather than work through another list of paper titles.

It addresses “I have read something relevant, but cannot remember where” and helps surface passages the writer may not yet know to look for. Passage relevance is not evidence that a claim is supported.

Guide: https://flowing.works/en/docs/recall

### Evidence Card: does the evidence support this claim?

After Recall has keywords and matching snippets, Flowing checks qualifying draft paragraphs against those passages. A left-gutter mark opens a dialog with the paragraph and underlined claims beside the evidence cards.

- **Red: contradiction.** The claim conflicts with the source in polarity, number, mechanism, or scope.
- **Green: direct support.** The source entails the same claim, polarity, and scope, without adding scientific inference. Same-topic overlap is not enough.
- **Yellow: useful qualifier or extension.** The evidence supplies a condition, boundary, limitation, or other relevant addition not yet in the paragraph. Yellow does not mean missing support.
- A card can rely on one or two complementary passages; a valid single-passage card is retained even when passages cannot be combined.
- Purple / blue keyword highlights belong to Recall, not a fourth verification verdict.
- **No card does not mean a claim passed verification.** Checks cover retrieved passages, not the entire library or public web. They do not produce “insufficient evidence” cards or rewrite the manuscript.

Guide: https://flowing.works/en/docs/evidence-verification

### Autocomplete: keep the next sentence moving

Click the sparkle beside the cursor or press Tab to preview the next sentence as gray inline text, using the nearby manuscript argument. No prompt or chat is required. When needed, guide the result with a note, selected library passage, or manuscript figure. Review the preview, then press Tab again to insert it.

When library content is used, matching source passages remain inspectable. “Evidence-backed” means source-transparent, not independently fact-checked or guaranteed correct. Autocomplete is the lightweight next-sentence path; Continue writing supports fuller discourse moves or evidence synthesis.

Guide: https://flowing.works/en/docs/autocomplete

## Why these capabilities belong together

**Manuscript → relevant passages → claim check → researcher judgment → writing.**

Reference evidence constrains what the text may responsibly say. Manuscript context helps determine what belongs here. Flowing's distinction is bringing both into the writing environment, not simply generating more text.

Supporting functions include Ask with context (manually attach passages, then ask), Polish (revise using context and evidence), and Continue writing (advance the argument at the cursor). All require researcher review.

## Getting started and tool fit

1. Import a few PDFs needed for the current section into a local library.
2. Write or import the draft you are working on and use its context to recall relevant passages.
3. Inspect sources and evidence cards, then preview a completion, ask, or revise.

- macOS and Windows. Free to start; no invitation required. The researcher program is an optional promotion.
- Rich-text and LaTeX editing, PDF preview, and Word (.docx) / PDF import and export. Check formatting for complex manuscripts.
- A references sidebar parses the manuscript bibliography; `@` inserts a citation from the library.
- Keep Zotero / Mendeley / Paperpile as the canonical reference manager and import the PDFs needed for writing. No direct Overleaf sync.
- Optional OpenAlex paper recommendations are off by default and separate from evidence verification. Enabling them sends keywords; paragraph text is sent only with opt-in.
- PDFs and the index stay local; the full library is not uploaded. AI features send needed manuscript context, matched passages, and manually attached content. Do not describe the product as fully offline or as transmitting only individually hand-picked excerpts.
- Full retention and model-provider training commitments are not yet published. Contact the team before using restricted or confidential material. Privacy: https://flowing.works/en/privacy
- Do not equate source transparency with zero hallucinations, exhaustive fact-checking, or automated academic judgment.

## Version and downloads

- **Version:** 0.1.40
- **Download (macOS, Apple Silicon):** https://cdn.flowing.works/flowing-release/macos/Flowing_0.1.40_aarch64.dmg
- **Download (macOS, Intel):** https://cdn.flowing.works/flowing-release/macos/Flowing_0.1.40_x64.dmg
- **Download (Windows x64):** https://cdn.flowing.works/flowing-release/Flowing_0.1.40_x64-setup.exe

## Pricing (monthly, USD)

- **Free:** $0 — 120K AI tokens/month; 2 Evidence Card checks/month; unlimited manual keyword search; local PDF library, editor, and References included
- **Go:** $5/mo — 1M tokens/month; 10 Evidence Card checks/month
- **Pro:** $10/mo — 4M tokens/month; automatic / continuous 🟢🔴🟡 Evidence Card
- **Pro+:** $30/mo — 16M tokens/month (4× Pro); strongest models; heavy research workflow

Prepaid token packs (1M $5 / 3M $12 / 10M $35) are available as backup after the monthly allowance. Web checkout starts in the desktop app (claim code), then the user picks a plan on the pricing page and pays with Stripe.

## Examples and credibility

- [Exploratory comparison at four manuscript positions](https://flowing.works/en/blog/ai-paper-continuation-comparison): a small product case study retaining first outputs, not a large blinded benchmark or proof of superiority over all chat models.
- [Evidence before fluency](https://flowing.works/en/blog/evidence-before-fluency): claim-source alignment. Correlation alone does not establish causation; a missing causal bridge does not itself warrant a yellow card.
- [Ness Labs founder interview](https://nesslabs.com/flowing-featured-tool): Featured Tool / Tools for Thought coverage, not an independent performance evaluation.
- [Community notes](https://flowing.works/en/community): source links and disclosures; personal experiences are not established universal product outcomes.

## Main entry points

- [Homepage](https://flowing.works/en)
- [Pricing and current allowances](https://flowing.works/en/pricing)
- [Step-by-step video tutorials](https://flowing.works/en/tutorial)
- [Full user documentation](https://flowing.works/en/docs)
- [FAQ](https://flowing.works/en/faq)
- [Fit alongside other tools](https://flowing.works/en/compare/zotero)
- [Support and contact](https://flowing.works/en/contact): support@flowing.works
- [Full text for AI agents](https://flowing.works/llms-full.txt)
- [Sitemap](https://flowing.works/sitemap.xml)