Clausehound turns statutes, regulations and contracts into a versioned clause model: git for the law. It runs where your data needs to live, and you connect from the AIs you already use and tokens you've paid for. No lock-in, and every answer comes back with its source.
Looking for lawyers, not software? Cobalt is the law firm that runs on Clausehound.
tools/call compare_statute_versions
{
"citation_code": "00e41", // Employment Standards Act, 2000 (ON)
"from_version": "2004-03-01",
"to_version": "2009-11-06"
}
← structuredContent
{
"changes": [
{
"change_type": "amended",
"section_path": "Part VII › 17",
"to_text": "…48 hours in a work week…"
},
{
"change_type": "added",
"section_path": "Part XVIII.1 › 74.1",
"to_text": "\"assignment employee\" means…"
}
]
}From your agent: two cited changes, not two consolidations of the Act.
query Findings($report: Uuid!) {
searchAnalysisWorkspace(searchQueryUuid: $report) {
verdict
rationale
approvalStatus
contractVersion { name contract { name } }
}
}
← data
{
"searchAnalysisWorkspace": [
{
"verdict": "inadequate",
"rationale": "No prior consent before customer
data is used to train AI models…",
"approvalStatus": "approved",
"contractVersion": {
"name": "Signed 2026-04-29",
"contract": { "name": "Master Services Agreement" }
}
}
]
}From your code: the findings your lawyers reviewed, with their sign-off.
A statute, a regulation and a supply agreement are the same thing to Clausehound: a document with versions, broken into a tree of clauses that can be tagged, compared and cited. Think of it as change control for legal documents: git for rules and regs. Your originals stay untouched; the model is a layer on top.
MCP is one door, not the only one. Here's what your end users see: search every document at once, compare any clause with every similar one, spot what's missing, and run a playbook of prompts with a person reviewing every result. The same findings reach your systems over GraphQL, and your agents over MCP.
Cobalt, a tech-focused law firm, runs its legal services on Clausehound: Diligence Monster for due diligence, DealPrep for deal knowledge and Policysaurus for policy research, plus records retention schedules, with its lawyers verifying every finding in the app. It's the platform's toughest user, and the reason every feature here was built for work that has to be right.
Want the same for your own legal team? If you're in-house counsel or run a legal team, we'll set up an engagement that puts the platform Cobalt uses in your hands: your documents, your playbooks, your deployment (behind your own firewall if you like), run by the engineers who built it. Set up an engagement →
You won't sign up, start small, and burn through your compute and AI budget in a week because the platform wastes tokens. The hard work happens once, at ingest. After that, a question is a lookup over structured clauses, not a model rereading documents from scratch. And the same design that keeps it cheap is what keeps answers grounded.
No shared Clausehound cloud: the data layer runs where you need it, hosted by us, in your cloud or on your own servers. Already approved your AI provider? Clausehound adds no new party holding your documents. UUIDs everywhere and WebAssembly builds keep it portable.
The core is Rust: compiled, memory-safe and lean. Answers come back in milliseconds, and a full deployment runs comfortably on modest servers. It's how we run it ourselves.
A tool call returns only the provisions that answer the question, with their sources. Your agent's context stays small, so every call costs less and runs faster than pasting whole Acts into a prompt.
A model handed three cited clauses has far less room to invent than one handed a 400-page PDF. Efficiency and accuracy come from the same place: structure.
Every amendment is a new, immutable version, chained to the last. Diff any two dates clause by clause, see what changed and why, and get flagged when a clause you rely on moves.
SOC 2 audited. PDF parsing and OCR run in their own service on an isolated network. Single sign-on through Azure AD and OpenID Connect.
AI is genuinely useful in legal work, and used carelessly it can hurt you. Language models are fluent and sometimes wrong, and in law a confident wrong answer is the expensive kind. We provide the platform, the tooling and the support to use AI responsibly.
"AI tools promise efficiency; however, their indiscriminate use in drafting legal arguments constitutes a pernicious threat to the integrity of the judicial process. Fabricated case citations distort and corrupt the case law."
Illustrative example. The flagged row cites a section that wasn't in force on the date asked about.
Most legal AI tries to one-shot answers from a model trained mostly on American law, and treats everywhere else as an edge case. Clausehound reads each jurisdiction's own official consolidations, with every historical version kept, so the answer comes from the law that actually applies. Our deepest corpus today is Canadian, because that's where we do the most work. The same pipeline takes on the EU, the UK or anywhere else that publishes its law.
For agents, Clausehound speaks MCP, the open standard AI assistants use to connect to tools. Point Claude, or any MCP client, at your endpoint and it can search, compare and cite the law directly. For everything else there's GraphQL and exports.
Clausehound isn't a thin layer over someone else's chatbot, built last year to chase a funding round. We've been structuring legal documents since 2016, doing machine learning on contracts with academic partners before anyone called it AI, and we patented the core. What we sell is what already works in production, not what's trending.
Clausehound starts: contracts as structured clauses, not files.
Joins Seneca HELIX, the start of a long research partnership with Seneca.
Machine learning on clauses with Seneca College: flagging problem clauses against a client's own precedent.
Patent filed for the clause taxonomy: breaking legal documents into typed, nested clauses that can be versioned, compared and analysed as data.
Large language models built into the platform, with people verifying every result.
Patented: Clause Taxonomy System and Method for Structured Document Construction and Analysis, US Pat. No. 12,670,200. Point-in-time law, MCP for agents, and Rust from end to end.
Chief Technology Officer · Co-inventor
Joshua leads Clausehound's engineering and co-invented its patented clause taxonomy. He has built the platform since 2016: its Rust services, its version model and its MCP server, with one rule for AI in legal work: every result tied to its source.
He started at IBM, inside the DB2 database engine and later as a web-security subject-matter expert, work that earned him a patent. In 2011 he founded Qaribou, building geodata, document-processing and fintech products for civic nonprofits and large enterprises alike. B.Eng., Western University.
More about Joshua
Founder · Co-inventor
Rajah is a legal technologist and a technology and M&A lawyer, trained at Stikeman Elliott. Before law he spent a decade in IT, as a developer and then an information systems manager. He founded Clausehound, co-invented its patented clause taxonomy, and leads Cobalt, the firm that runs on it.
M.B.A. and J.D., Western University (Ivey and Western Law).
With engineers Ian Hume and Isaiah Liu.
Licensing the platform, connecting your stack, or an engagement where our engineers set it up and run it for your legal team: we'll walk you through the app, the API and the MCP server, and talk about a deployment of your own, including on your own infrastructure.
Already a customer? Log in. Or email [email protected]. Need a lawyer? Talk to Cobalt.