Alexi vs Caseway: how they compare in 2026

Alexi describes itself as a private single tenant AI engine, built originally to generate evidence backed research memos, with a proprietary model layer and a retrieval first architecture. Memo generation, precedent finding, chronology building and workflow automation are all model driven.

Casey is a generative research assistant that answers legal questions in conversation with answers cited to Canadian and United States case law, plus analysis of uploaded documents linked to citations. CaseForm completes court and regulatory forms by extracting required fields from source documents and prior filings and applying rules specific to each jurisdiction before submission. Synthium is a retrieval and validation layer for complex regulated work. The agreement defines Output as material the platform generates from Input, acknowledges that the platform uses probabilistic prediction and pattern analysis with inherent limits, and treats third party large language models as a dependency. CaseLite, a free keyword and citation search over the same collection, does not use the models; the vendor positions it as the starting point from which a user upgrades to Casey for AI answers.

Alexi submitted to the 2025 Vals Legal AI Report legal research study, run by an outside evaluator that published its rubric and weights in advance: accuracy 50 percent, authoritativeness 40 percent and appropriateness 10 percent, across 200 US legal research questions from attorneys at named firms. In the evaluator's published results, Alexi scored a 77 percent weighted aggregate, second of the four systems measured, against a lawyer baseline of 69 percent. The vendor's own account leads with 80 percent overall accuracy against a 71 percent lawyer baseline, and 79 percent accuracy on single jurisdiction questions described as tied for the highest, and does not state that it placed second on the weighted aggregate. Several major vendors declined to take part and at least one withheld permission to publish. All systems measured struggled on questions spanning several jurisdictions and trailed a generalist model on fifty state statutory surveys. The architecture is described as retrieval first, to ground output in primary law rather than model memory.

Answers are cited to real court decisions, and the methodology page states that the product does not generate opinion text or fabricate citations. Opening a result shows structured metadata and documents tied to a real source identifier from the upstream publisher. The same collection is free through CaseLite without an account, and the vendor invites users to run a known citation and see how the record appears. The methodology page states that not every court, tribunal or period is equally represented, that absence of a case does not mean it does not exist and that older records may lack citations or dockets, and it tells users to confirm critical citations against the official reporter when filing. The agreement tells users always to confirm that cited sources support the statements made. No accuracy figure, error rate, test set, benchmark or evaluation is published.

The product runs workflows of several steps that carry legal principles and citations through successive stages, and the vendor states that consistent, verifiable accuracy is what makes that automation safe. The CEO has spoken publicly about human oversight in the context of a long term vision for AI assisted arbitration. The vendor site, FAQ, workflow and product pages and blog describe nothing about what a workflow does unaided, a threshold at which it stops or escalates, a review surface given to a lawyer, or what the vendor commits to when an output is wrong.

The agreement says the user alone must review and verify any Output before using it in a legal matter, that outputs may be incorrect or incomplete, that Output should not be relied on without advice from a licensed attorney, and that the vendor is not liable for actions taken on it. The methodology page repeats that search is a research aid rather than legal advice. Nothing published describes what runs without a person, when the system acts on its own, what a review step looks like in the product, or whether any human checkpoint exists. CaseForm applies validation specific to each jurisdiction before submission of court and regulatory forms, and nothing states who presses submit, what the validation checks, or what happens when it passes something wrong. No confidence signal, escalation path, audit view or reviewer workflow is described for any product.

The vendor states it is trusted by leading law firms and names buyer roles: CIOs and CTOs, knowledge management leaders, innovation teams, practice group leaders and managing partners. The vendor site, blog, FAQ and news announcements name no customer and publish no case study pairing an organization with figures and a date. The quantified material is benchmark performance: the Vals figures measure the system on a test set rather than at a firm that deployed it.

A research collaboration with the University of British Columbia is reported in the legal trade press, and the vendor describes it as anchoring AI systems in real court decisions. CaseForm has a named integration with MyCase, a practice management product. The free search can be used by anyone without an account, and trade coverage runs across Canadian Lawyer, CBC and regional technology press from 2024 to 2026. The pricing, methodology and product pages name no customer, quantify no deployment, and publish no time saving, cost saving, user count or matter volume, and the methodology page tells users to treat size claims as a point in time. The site navigation also lists a Success Stories page and a Testimonials page.

The FAQ states that nothing leaves the firm's private environment and that a firm's prompts, documents and workflow data never train global models and never mix with other firms, and a document management partner's listing separately records no training on customer data. Every firm on the private cloud model receives a fully isolated single tenant instance, and the vendor lists what is isolated: retrieval, workflows, memory, security protocols and governance. Memory is on that list, so one firm's accumulated intelligence is not carried toward another. Supporting controls are AES-256 encryption, TLS 1.2 or higher, single sign on, role based access control and audit logs, under SOC 2 certification. No retention period or deletion commitment is published.

Clause 7 makes customer data the customer's confidential information and binds each party to use the other's only for the agreement. Disclosure is limited to employees and contractors who need to know under equal obligations, with at least reasonable care. Around it sit a zero retention default, deletion of uploads within twenty four hours, chat history saved only on opt in and deletable at any time, and a bar on training public models on customer input. Clause 3 states that because the user is not receiving legal advice, nothing entered is protected by attorney-client privilege or the work product doctrine, and tells users not to enter confidential or sensitive information unless they are entitled to and accept that it will not be privileged. Nothing addresses separation between matters or between users within one customer.

The audience is lawyers and law firms, with named buyer roles inside firms and no consumer or non lawyer surface. The vendor engages publicly with the problem of hallucination in filings and treats accuracy as a question of professional risk. The vendor site, FAQ, blog and product pages publish no position on the advice line, on competence or supervision duties, or on jurisdiction limits, and the product spans Canadian and US law, where the professional rules differ.

The agreement says the platform is not a law firm and gives no legal advice, and that using it creates no attorney-client relationship. Output is general information generated by AI, not a substitute for professional counsel, and users agree not to rely on it without advice from a licensed attorney in the relevant jurisdiction. The vendor disclaims responsibility for actions taken on it, and the warranty section and the methodology page repeat the point. A jurisdictional caveat says the platform may not be appropriate or available everywhere and that users where it imposes extra obligations proceed at their own risk. The product is sold to the public as well as to legal professionals. No provincial law society rule, statutory prohibition on unauthorized practice, bar guidance or court rule is named, in a market where the rules are made province by province.

Alexi submitted to an independent third party evaluation and published the results, and argues publicly that transparency and accountability in legal AI require that rather than marketing claims. The evaluation covers one point in time. The vendor site, FAQ, blog and news announcements publish no AI governance framework, AI management certification such as ISO 42001, named owner of model governance or testing gate before release, and nothing addresses uneven output across matter types, parties or populations.

The agreement states that the platform's AI involves probabilistic predictions and pattern analysis with inherent limits, and that the vendor does not guarantee that Output will be correct, complete or suitable, or that any legal result will follow. The methodology page publishes five named limitations of the collection, including uneven representation across courts and historical periods and the warning that absence of a case does not mean it does not exist. The privacy policy commits that legal content is not used for model training, profiling or advertising. No AI governance policy, framework or standard is published, and no certification such as ISO 42001 is claimed. Nobody is named as accountable for the system's behavior, and no evaluation before release, red teaming, benchmark or monitoring is described. No bias, fairness or representativeness disclosure is published; the product is built on judicial decisions about identifiable parties, and its methodology page states that representation is uneven.

The security page and FAQ list AES-256 encryption at rest, TLS 1.2 or higher in transit, single sign on, role based access control, audit logs, support for private VPC deployment, and deployments in specific regions for firms operating across offices. Each firm gets a fully isolated single tenant instance covering retrieval, workflows, memory, security protocols and governance. SOC 2 certification underpins the control set, and a Trust Center offers a request route to a security addendum, architecture whitepaper, data processing agreement and audit reports. The security page and FAQ state no retention period or deletion control for research queries, uploaded case files and generated memos, no named subprocessor list, and no incident or breach notification practice.

Clause 7 of the agreement sets a zero retention default: session content, uploads, queries and prompts are stored only temporarily to generate Output unless the user saves them. Uploaded documents and context files are deleted within twenty four hours, chat histories are saved only on opt in and can be deleted at any time, and the form product keeps no case or matter data on the vendor's servers. The vendor will not train public AI models on customer input, and the privacy policy says legal content is not used for model training at all, with no sale of personal data and no advertising profiles. Security is described generally as encryption in transit and at rest, access controls and monitoring. No subprocessor or service provider is named, and no incident or breach notification practice, timeline or security contact is published. The agreement states that the vendor is not HIPAA or PCI DSS certified and tells users not to process health or payment card data.

No indemnity, liability cap, carve out, warranty on output or insurance position is published, and no customer terms of service or master agreement is published on the vendor's site. The vendor positions itself on measured accuracy and on reducing the risk of hallucinated citations reaching a filing.

Clause 12.1 commits the vendor to defend the customer against a third party claim that the platform, used under the terms, infringes a valid copyright, patent or trademark or misappropriates a trade secret. It indemnifies against damages, costs and reasonable fees awarded or settled, with remedies of modifying the platform, obtaining a license, or terminating with a pro rata refund. Liability is capped mutually at the greater of twelve months of fees paid or CAD 500, with fraud, gross negligence and willful misconduct capped at twice that. Termination for convenience by the vendor carries thirty days notice and a pro rata refund of prepaid fees. The platform and all content are provided as is, and the accuracy or reliability of Output is expressly not warranted. No service level or insurance position is published. Clause 8 disclaims liability for failures of third party AI models.

A product integration with iManage is announced. The vendor site, FAQ, product pages and news announcements publish no integrations index, name no other connector for practice management, court filing or Microsoft Word, and do not document what the iManage integration moves, in which direction, or what an administrator configures.

The form product integrates with MyCase, a practice management platform, on its own page in the site navigation. API access is stated for the research assistant for research and document workflows, and API or embedded interface options are offered for the form product. For the enterprise data hub the vendor describes custom pipelines and integrations tailored to a customer's systems. The pricing and product pages link no API or developer documentation and describe no authentication or scope model. No document management system, email client, word processor or electronic signature product is named, and nothing says what moves in which direction across the MyCase connection or how permissions carry over. The site navigation also lists an Integrations and Partnership page.

Two models are published. A private single tenant environment per firm is the default, and private cloud deployment is offered for firms wanting complete ownership and control of their own data. Single tenancy is stated as architecture rather than as an enterprise upsell. The vendor site, FAQ and product pages name no regions, offer no customer selectable residency, and do not say where processing happens as distinct from where data is stored. Alexi operates across Canadian and US jurisdictions.

For the research assistant, no country or region is named for storage or processing, no residency option is offered, no tenancy model is described and no cloud provider is identified. The privacy policy says what is kept and for how long but never where, and the agreement's only geographic content is a governing law clause and an arbitration seat. For the enterprise data hub, the vendor describes secure deployments with data residency options and compliance support, on premises or in an isolated cloud. Nothing connects that to the research assistant a lawyer buys at the published seat price, and nothing says the assistant can run on premises, in isolated tenancy or in a chosen region.

The vendor publishes a dedicated security page stating SOC 2 certification and operates a trust center at trust.alexi.com, which lists SOC 2 Type I and SOC 2 Type II reports, a 2025 penetration test, a privacy compliance policy and security policies for cryptography, access control, incident response, operations and continuity, each released through a request access flow without a sales call. The security page says the trust center also carries a security addendum, an architecture whitepaper and a data processing agreement, which its resource list does not show. SOC 2 certification is also stated on the vendor's platform pages, about page and FAQ, and in a document management partner's technology listing. Supporting controls published on the FAQ: AES-256 encryption, TLS 1.2 or higher, single sign on, role based access control, audit logs and support for private VPC deployment. No coverage period, audit scope, report date or named auditing firm is published.

No certification is claimed and no trust center exists. There is no SOC 2, ISO 27001, independent audit, named auditor, penetration test result, compliance page or portal for requesting documents. The site navigation has no security or trust entry; its legal section lists only the privacy policy and the terms. The agreement states that the vendor is not HIPAA or PCI DSS certified and tells users not to process health or payment card data, and no unsupported badges appear on the site. Clause 7 refers to a Statement of Confidentiality with more technical and legal detail on how customer data is protected; no location is given for it, it does not appear in the navigation, and it is not published.

A proprietary model layer the vendor brands as Alex is published, described as trained specifically on law rather than general purpose, with a retrieval first architecture. The trust center's subprocessor list names Microsoft Azure in Canada as cloud provider, Weaviate as the vector database for embedded public legislation and Heroku as a platform, and names no model provider. The vendor site, FAQ, product pages and blog name no foundation model or provider underneath the proprietary layer, say nothing about where models run, and make no commitment to notify customers when the supply chain changes.

Clause 8 of the agreement says the platform may rely on third party services, including third party large language models or data sources, and that some customer input may be processed by third party AI engines under the vendor's contracts with them, so users are told that outside models see their material. No model, model family, provider, inference host or cloud provider is named anywhere, including in the privacy policy, which refers only to contracted service providers acting on instructions. The same clause reserves the right to replace, update or disable third party services at any time, and says a change to or loss of a third party AI model gives the customer no right to a refund or relief. The vendor commits that third party processing will be under appropriate confidentiality and data protection obligations and disclaims liability for those providers.

No pricing page, rate, unit of charge or tier structure is published. Every commercial path ends in a demo or contact request. No free trial or self serve entry point is offered, and no third party pricing figure is published either.

The research assistant costs CA$49.00 per month, with a monthly and yearly toggle. The form product costs $49 per seat per month with its own purchase path, and the enterprise data hub has a contact route. Search is offered free with no account, and the vendor describes the upgrade from it to the AI assistant. Students get fifty percent off and veterans fifteen percent, with eligibility checks. The agreement sets fees in Canadian dollars excluding taxes, billed monthly or annually by card or invoice, with late interest of 1.5 percent a month. Fees are non cancellable and non refundable, with no refund for cancellation mid term, a thirty day window to dispute a charge, and a pro rata refund where the vendor terminates for convenience on thirty days notice. No usage limits, quotas or overage terms are published. The assistant's feature list refers to uploads for premium users while only one price is shown, so a premium tier exists without a published price.

The buyer is law firms, and the vendor names the roles inside them: CIOs and CTOs, knowledge management leaders, innovation teams, practice group leaders and managing partners. Practice coverage spans litigation, transactional, operational and administrative work, with a published treatment of personal injury covering case file summarization, chronology building and memo drafting. Jurisdictional scope is Canada and the United States. Nothing states which firm sizes or practice areas the product is not built for, and no coverage is listed at state or provincial level.

Case law covers the United States, federal and state, and Canada, federal, provincial and territorial. Primary law browsing covers California, Florida, New York and Texas, plus Canadian federal, provincial and territorial collections as released. Records carry party names, courts, decision dates, docket or court file identifiers and citations where the source has them. Ingest is ongoing as sources publish, and the vendor says depth and lag vary by court and recent filings may appear after a delay. Published limits: coverage of courts, tribunals and periods is uneven, absence of a case does not mean it does not exist, older records may be incomplete, and primary law outside the listed jurisdictions may be missing. Buyers named are law firms, enterprises and individuals, with legal alongside defense, manufacturing, energy, mining and utilities. No depth by court, date range or historical start point is given for any collection.