heidihelena/citevahti

CiteVahti

Check every claim before you cite it.

CiteVahti tests whether a manuscript claim is supported by the paper cited for it. You rate first. AI gives a blinded second opinion. You adjudicate. The Zotero write-back is previewed, confirmed, audited, and undoable.

A product of Vahtian. Free and local-first for researchers; Vahtian sells paid infrastructure to organizations that need auditable citation integrity at publication scale. The open Apache-2.0 core never paywalls a researcher's ability to verify their own manuscript.

The human or panel is always the decider. The AI is a blinded, advisory second rater only — its values are advisory, never decisive, and never silently propagated.

Choose your path

I write manuscripts

Use the browser panel — no terminal required. Paste a paragraph, rate whether each cited paper supports its claim, and verify your first claim in ~10 minutes. → Getting started · docs/QUICKSTART.md

I use agents or VS Code

Install the MCP server and run claim tests from Claude, ChatGPT, Codex, or the VS Code inline review loop. → Path A — chat-driven

I review manuscripts for a journal

Export an auditable claim-evidence trail for methods reporting, then install. → docs/REPORTING.md

Beta. Free to use for testing, research feedback, and early development. Pricing for hosted and advanced features may come later; a free local/community version is intended to remain available.

Full status & capabilities — what's complete in v0.16.0, the two co-primary surfaces, the literature sources, and the VS Code adapter: docs/STATUS.md.

See it

The journey, in three screens: paste a paragraph, rate each claim against the paper cited for it, and let the AI's second opinion appear only after you've rated.

![CiteVahti highlights c