We built the associate we would have wanted to supervise.

One that shows its working, never cites something you cannot open, and hands the decision back to the person whose name goes on the letter.

Book a 20-minute walkthrough Start free

A product of Cognesio LLP

The How this was made tab: the plan, the sub-agents that ran, the checks that passed and the redactions applied before any model saw the prompt. — The tab we are proudest of: How this was made. The plan, the checks that ran, and what was redacted before a model saw the prompt.

A citation is a promise.

When a lawyer cites an authority, they are promising a judge, a regulator or a client that the thing exists and says what they say it says. General AI tools break that promise quietly, because a plausible citation and a real one look identical on the page.

That is not a hypothetical, and it is not confined to one country. In England and Wales, for example, the SRA raised it in its Risk Outlook on artificial intelligence in November 2023. In June 2025 the Divisional Court dealt with Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank, on fabricated authorities put before the court. The SRA's thematic review of compliance officers followed in December 2025, and its guidance update on new technology and COLP responsibility in February 2026. In the United States, Mata v Avianca (S.D.N.Y. 2023) sanctioned lawyers for citing cases that did not exist, and a public tracker maintained by Damien Charlotin now counts such cases in courts around the world.

The lesson we took from all of it: a legal AI product's most important feature is what it refuses to do.

Read the verification page

What the product refuses to do.

Each of these makes the product worse at something and much better at the thing a supervising lawyer needs. We took every one of the trades.

01
Approved sources, not the open web.
Every cited passage is checked against a fixed list of approved public sources. That list covers the United Kingdom, Ireland, the European Union and the European Court of Human Rights today, alongside international bodies. Where a matter sits outside it, research runs at open-web quality with the same gates and the output says so. More jurisdictions are added as sources are certified.
02
No licensed content.
Westlaw, Practical Law, LexisNexis, Bloomberg Law, Justis and vLex are refused outright. It means we cannot match a competitor's corpus. It also means we never cite something you cannot open, and never push your client's document into a subscription you have not licensed.
03
Deterministic checks, not a second model.
Four gates run over the work of every configured agent, using set membership and string matching rather than judgement. A model marking a model's homework is not a check.
04
A document, not a conversation.
What comes back is shaped for the person who has to read it: the board, the fee-earner team, the client, the court, the regulator. Chat is how you ask. A memo is what you get.
05
The matter, not the folder.
Client, reference, responsible partner, jurisdiction, key dates and parties, with a conflict check when the matter opens. Everything attaches to the file.

Position, plainly.

Harvey sells to large enterprises, US-led and document-management-first. Legora is strong on table review and collaboration. CoCounsel and Lexis+ AI are built on the corpora their parents own. Copilot and ChatGPT are general tools rather than legal ones.

They are all serious products. Ours is a different shape: verification you can inspect line by line, work shaped by audience, governance that lives on the matter, and a price a firm of twenty lawyers can sign without a procurement process.

If your firm's requirement is a licensed US corpus or a live iManage integration today, we are not the right answer yet. We would rather tell you that in the first meeting.

See how it compares

Roadmap lines are labelled as roadmap lines.

The product ships behind one rule that also governs this website: nothing is described as working until it works. The marketplace lists DMS tiles that are not yet live integrations, and we say so on the security page rather than leaving the tile to imply it. The printable bundle carries a content digest, and we call it a content digest rather than a hash chain.

You will find that rule applied unevenly by our competitors. You will find us boring about it.

Read the changelog

Firms of five to a few hundred lawyers, wherever they practise.

Firms that answer to a regulator, that do not have an innovation department, and that cannot spend six months on a procurement exercise to find out whether AI helps. A matter carries its own jurisdiction, so the same firm can run English, Irish, EU and cross-border work side by side.

In-house legal teams of ten to fifty have the same shape, and the product fits them without modification.

Built by Cognesio LLP.

LegalAI Space is a private AI workspace for legal work, built by Cognesio LLP. The team has built software for regulated professions before, and the product carries that habit: an audit register that projects from the operational record, a share link that hashes its passcode, a bundle that says content digest rather than blockchain.

We are a small team and we answer our own email. If a page on this site says something the product does not do, tell us and we will change the page, not the story.

Three lawyers around a table of contracts, one pointing at a clause.

Two lawyers walking and talking outside a Georgian brick building.

A general counsel presenting to colleagues in a boardroom at dusk.

Legal entity: Cognesio LLP
Write to us: hello@legalaispace.com; We answer in within one business day.
See it working: Book a walkthrough; Twenty minutes, on a document you choose.