Alexi

Sweet spot: a 5–50-lawyer litigation boutique where associates spend a meaningful slice of their week on research memos, the firm wants AI but does not want to bet on a heavyweight platform play, and a single-tenant deployment is appealing for client-confidentiality reasons. Alexi compresses memo turnaround from a day to under an hour and gives senior litigators a credible draft to mark up rather than a blank page. Failure modes. Cite-verify every output — the VALS rating is real but it is not zero hallucination, and the cost of one fabricated case getting into a brief is measured in sanctions and reputational damage. Treat Alexi as the first-pass researcher, not the citator. Source-corpus depth on niche or older case law trails Lexis / Westlaw, so it is best paired with an incumbent for Shepard's / KeyCite-style validation rather than used as a sole-source provider. Model bias toward common-law commercial reasoning means highly specialised areas (admiralty, ITC, niche regulatory) need more verification. What to pilot. Pick 10 recent research questions where you know the right answer. Run them through Alexi. Score for accuracy of cited authorities, completeness of analysis, and time saved versus the associate's baseline. Above 80% accuracy with full cite-verification still adding value, the seat math works for any litigation-heavy practice; below that, expand the test set before scaling.

Alexi is a Toronto-founded legal-AI platform focused on the research-and-memo workflow that litigators run dozens of times per matter. You ask a legal question — "in Ontario, what is the standard for striking a pleading as an abuse of process?" — and Alexi returns a structured research memo: applicable law, leading cases with summaries, distinguishing authorities, and an answer-the-question section. Drafting and document-review modules layer on top: compare two shareholder agreements clause by clause, generate hearing-readiness checklists, draft initial litigation correspondence. Alexi's core AI pipeline is a research-question pipeline plus memo generator: the question is decomposed, jurisdiction is detected or specified, primary and secondary sources are retrieved, and the model writes a memo that cites only the retrieved authorities. The "active learning" framing means the system improves with use inside a firm's tenancy — single-tenant private deployment is offered for firms that need it. Independent benchmarking from VALS AI has rated Alexi at the top on legal-AI accuracy in their published evaluations, which the company leans on heavily as a differentiator. Competitively, Alexi is the pure-play legal-research pick. Lexis+ AI is the incumbent — bigger source corpus, deeper Shepard's citation validation, broader product surface, but historically more conservative AI features and tied to a Lexis subscription. Casetext CoCounsel (now part of Thomson Reuters) is the other pure-play, deeply integrated into Westlaw, with strong brief-analysis features. Alexi differentiates by being lighter weight, faster to onboard, jurisdictionally strong in Canada (a real gap for Lexis+ AI / Westlaw equivalents historically), and willing to ship single-tenant deployment for risk-averse firms. For US litigators who already have a Lexis or Westlaw subscription, the pitch is "Alexi answers research questions faster and writes the memo — keep your incumbent for citation validation." Pricing is sales-gated; Alexi runs a guided trial / consultation funnel rather than self-serve checkout.