Drop legal docs.
Get a report.
Nothing leaves your browser.
Vaulytica is a deterministic linter for contracts. It runs hundreds of identical checks against your document, the boring exhaustive ones a tired human misses at 1 a.m., and hands back a Word file you can cite. Free. Open source. MIT. No account, no AI, no server. Drop something on me.
Multiple files? Drop a folder or
What I check.
115 rules at launch across ten categories — 220 more in v3, 730 more in v4 (1,000+ in total). Every finding cites the rule by id; legal assertions cite a statute or regulator by URL, drafting-practice findings cite their practice source.
Structural & formal
Party blocks, signatures, defined terms, cross-references, numbering, template placeholders.
STRUCT-001STRUCT-003STRUCT-005STRUCT-007STRUCT-009STRUCT-011
Financial terms
Word/numeral mismatches, currency consistency, usury, payment terms, MFN, caps.
FIN-001FIN-002FIN-003FIN-004FIN-005FIN-007
Temporal
Impossible dates, auto-renewal notice windows, survival, cure periods, post-expiry references.
TEMP-001TEMP-004TEMP-005TEMP-006TEMP-008TEMP-010
Risk allocation
Indemnity presence + mutuality + caps, LoL with exceptions, consequential waivers, insurance.
RISK-001RISK-002RISK-004RISK-005RISK-009RISK-013
Choice & venue
Governing law, forum, foreign-venue enforceability, arbitration, class waivers, jury waivers.
CHOICE-001CHOICE-004CHOICE-005CHOICE-006CHOICE-007CHOICE-008
Dark patterns
Unilateral modification rights, buried auto-renewal notices, asymmetric fee-shifting, forced arbitration.
DARK-001DARK-002DARK-003DARK-004
How this works.
Your file goes in on the left, runs through a deterministic rule engine fed by the Knowledge Base, comes back out on the right as a Word document. Nothing else moves.
-
Your filePDF or DOCX, dropped in the tab
-
Read & extractparties, dates, clauses
-
Knowledge Basestatutes · clauses · playbooks · dark patterns
-
1,000+ cited rulesdeterministic engine
-
Word report.docx download
Everything runs in your browser — nothing leaves the tab.
What I do not do.
- I do not give you legal advice. I am a software tool. If something here matters, hire a lawyer.
- I do not replace your judgment. I find mechanical things consistently. The hard calls are still yours.
- I do not use AI. Not a model, not a copilot, not "powered by." A probabilistic answer cannot be cited. The whole point of this is the opposite.
- I do not see your data. There is no server. Your contract never leaves the tab. Open DevTools if you want to confirm: the only requests you will see are same-origin GETs of my own static assets — rule data, playbooks, code — and none of them carries a byte of your document.
Why no AI?
Every other contract tool you can find right now leans on a language model. The pitch is the same everywhere: paste your contract, the AI summarizes, the AI flags risk, the AI suggests redlines. The output is fluent and confident and changes every time you run it. A junior associate can use that for research. A senior partner cannot sign off on it. An auditor cannot trace it. A client cannot reproduce it. The output is, by construction, uncitable.
Vaulytica is for the work the AI tools cannot do: exhaustive, identical, cited checking. The professional failure mode in contract review is not "I did not understand the indemnity clause." It is "I read eighty pages at midnight and missed that the auto-renewal triggers sixty days before expiry rather than thirty." A deterministic checklist runner catches that failure mode every time. A chatbot is neither exhaustive nor reproducible. If you want to chat with an AI about your contract, do; it is fine, just do not do it here.
Your privacy.
Vaulytica runs entirely in your browser. There is no server to send your file to, because there is no server. Open DevTools, open the Network tab, drop your contract, and confirm it: the only requests are same-origin fetches of the app's own static assets — rule data, playbooks, code — and not one of them carries a byte of your document. Verify it yourself. The site is a static page on Cloudflare Pages, the code is on GitHub under MIT, and there is no analytics, no telemetry, no error reporting, no fonts from CDNs. Every byte your browser fetches comes from vaulytica.com.
Verify in DevTools · 30 seconds
- Open this page in Chrome, Firefox, or Safari.
- Press F12 (or ⌥⌘I on macOS) to open DevTools.
- Click the Network tab and check "Disable cache."
- Reload the page once to capture initial asset loads, then click Clear.
- Drop a contract onto the drop zone. Watch the network tab during analysis. You may see the app fetch its own rule data and playbooks from this same origin — every request is a GET with no body, and nothing from your document appears in any of them.
For court and ethics compliance.
Courts increasingly regulate generative-AI use in filings: the Ropes & Gray AI Court Order Tracker lists 300+ judge- and court-level directives, and the Legal AI Governance tracker counts roughly 113 orders that bind attorney filings — most requiring disclosure of AI use, verification of AI output, or a certification that neither occurred (New York's court-system-wide 22 NYCRR Part 161 took effect June 1, 2026). Privilege case law on consumer AI tools is diverging, not settled — compare United States v. Heppner (S.D.N.Y. 2026, no privilege for material run through a public generative-AI tool) with Warner v. Gilbarco (E.D. Mich. 2026, no waiver) — and ABA Formal Opinion 512 puts the verification duty on the lawyer.
Vaulytica can make the no-generative-AI certification provable rather than asserted:
every report can ship with a one-page verification certificate
(Word + JSON) stating the engine and knowledge-base versions, the input's SHA-256,
the result hash, and that the analysis was a deterministic rule evaluation with no
generative-AI, machine-learning, or probabilistic component — performed locally, with no
document content transmitted. The certificate carries its own tamper-evident hash, and
anyone can re-run vaulytica verify <report> <file> to reproduce
the analysis exactly. It certifies what this tool did — never your overall compliance;
the reviewing attorney remains responsible for verifying findings and meeting the
applicable orders and professional-conduct rules.
Where the rules come from.
Every rule cites at least one of these. Every source is free and public.
SEC EDGAR public domain
Real-world clause corpus sampled from EX-10 attachments to 10-K, 10-Q, 8-K, and S-1 filings.
efts.sec.govCornell LII free use
Cross-references between the UCC and each state's enacted version, and state-code lookups.
law.cornell.eduCommon Paper CC BY 4.0
Canonical balanced-default reference clauses for NDA, SaaS, MSA, SOW, DPA, and more.
commonpaper.comCUAD CC BY 4.0
510 contracts, 13,000+ expert annotations, 41 clause categories. Trains the deterministic classifier.
atticusprojectai.orgLEDGAR CC BY 4.0
80,000+ provisions across 100 clause categories. Broader training set for the classifier.
huggingface.co/lex_glueUS Code (LRC) public domain
Official US Code in USLM XML: Titles 9, 11, 15, 17, 18, 28, 35, and 41.
uscode.house.goveCFR public domain
Electronic Code of Federal Regulations: FTC (16), CFTC (17), IRS (26), Labor (29), HIPAA (45).
ecfr.govgovinfo public domain
Bulk Federal Register, CFR, Public Laws, and Statute Compilations in XML.
govinfo.govUniform Law Commission free use
UETA, UCC, UTSA, RUFADAA, and other uniform-act citations.
uniformlaws.orgawesome-legal CC0
Curated set of public legal templates: GitHub's BEIPA, Cooley offer letters, YC SAFEs.
github.com/ankane/awesome-legalAtticus Project CC BY 4.0
Clause-type definitions and the broader Open Contract Dataset.
atticusprojectai.orgOpen-Agreements MIT + CC BY
Fillable DOCX builds of Common Paper, Bonterms, and OpenAgreements templates.
github.com/open-agreementsFAQ.
Is this legal advice?
Can I trust the report?
Do you train models on my data?
Why no AI?
How is the data kept current?
What contract types are supported?
What if I find a bug?
Can I add a rule?
docs/adding-a-rule.md in the repository. Every new rule must cite
at least one DKB entry by id, include a positive and negative test fixture, and pass the
deterministic regression check.
Why MIT license?
What about other languages?
Stop hoping you didn't miss anything.
Prove you didn't.