Jhana.ai vs Lexroom: how they compare in 2026

The models are the product, and the archive was built to feed them. The vendor calls itself a frontier legal intelligence lab and names its proprietary work as verifier models, graph models for legal ontologies, a national legal archive of more than 16 million machine enhanced documents, and research and drafting agents. Machine enhanced is the key phrase: the archive was processed into a model asset, not licensed and wrapped, and seed capital was raised to build proprietary datasets and models and to hire researchers in law and AI. Every product is an agent pipeline, including the public sector agents for extraction, synopsis and forensic scrutiny. Without the models there is no product.

Lexroom's Libreria Lexroom is a curated collection of institutional legal sources organized into more than fifteen subject modules, each selected by legal institute and validated by partner jurists, and the commercial unit is access to those modules: the published entry plan buys one Lexroom module for one user. A curated legal database is a product that functions without any model, and it is what traditional Italian banche dati sell. Lexroom also states plainly that it does not build its own large language model but uses third party AI. The AI is the mechanism the buyer pays for, layered on a document collection that would still exist without it.

Grounding is built into the design and can be inspected, without any published measurement. The vendor says its agents always cite their work and names a verifier model as proprietary technology, putting a checking step in the design. Product material shows each conclusion one click from its source page and each extracted figure anchored to page and paragraph, which a reader can check. Nothing measured is published: no accuracy figure, retrieval or extraction precision or recall, hallucination rate, test set, independent benchmark, or account of what the verifier catches. For a vendor whose named innovation is a verifier, the missing verification result is the gap.

Every answer is anchored to updated sources that can be verified and downloaded in one click, and every statement links to the original source so the user can open, verify and cite it, according to the research page. Source selection is described: thousands of legislative and case law sources, selected by legal institute and validated by partner jurists, across more than fifteen subject modules built and verified one at a time. The privacy policy adds that Lexroom always provides the sources that justify the reasoning so the customer can check it. No accuracy figure, test set, benchmark or evaluation is published. The marketing makes an absolute claim in two places: zero hallucinations and zero risks, and on the research page, sources consulted without risk of hallucination.

The positioning implies an oversight model and no document describes one. Calling the product a paralegal rather than a lawyer is a deliberate and meaningful choice: a paralegal works under supervision, and the naming carries that. But naming is not a published model. Nothing states where a human must review agent output before it is used, whether any step can complete without review, what the agents may do unattended in the Courtroom and PUBSEC pipelines running extraction, synopsis and forensic scrutiny for courts, or what happens when an agent is wrong inside a registry workflow. No confidence threshold, escalation path or human confirmation requirement is published. The gap matters more here than for a research tool sold only to private practice, because the deployed users include judges and registrars.

Automated decision making under the GDPR is addressed directly in Lexroom's privacy policy. It states that outputs are always submitted to a human operator who must review them before use, so they are not automated decisions in the strict sense. Sources sit behind every statement so a user can verify before relying, the research page says the user decides rather than an algorithm, and the Word add in proposes changes inside the document rather than applying them. Nothing states what any step completes without a human, what the system does when it is uncertain, or what happens after it is wrong.

Adoption is quantified and institutional. The vendor reports more than 10,000 users, over 150 judges and registrars in five or more courts, more than 100,000 paralegal sessions and over 200,000 searches. Judicial use, with a public authority using the product rather than appearing as a logo, is the strongest element. Also dated and attributable: a $1.6m seed round led by Together Fund with named participants, a January 2026 partnership with CADRE ODR for online arbitration, and support from Jio GenNext, the AWS Public Sector Startup Hub, Microsoft Founders Hub and the Google Cloud Startups Program. Three gaps remain: every figure is self reported with no method, period or definition; no court, institution or firm is named; and no outcome measure is published, such as time saved, fewer errors or research quality, only volume.

Corporate customers include Satispay, Fastweb, Italgas, Mediolanum, CRIF, WST, MDV and Generali, and law firm customers include Withers, LCA and Gatti Pavesi. Four individuals speak on the record with name, role and employer, including Maddalena Malzanni, Legal Counsel Lead at Qonto, and Ferdinando de Martinis, Associate at Gitti and Partners. There are dedicated customer story pages for Qonto, Jet HR and Credem Banca. Figures are published separately from the customers: 40 percent time saved on repetitive tasks, 2.5 times more documents drafted, and more than 15,000 legal professionals using the platform. Those are aggregate claims with no method, basis or attribution to any named firm, and nothing is dated.

Real documentation exists, none of it about privilege, and one disclosed practice cuts against confidentiality. The privacy policy is a product document framed against the Digital Personal Data Protection Act 2023 and the Information Technology Act 2000, with erasure on request by email, retention exceptions for fraud prevention and for tax, legal reporting and audit duties, and a security section. Nothing addresses legal professional privilege or work product, the confidentiality of case files uploaded to AI Paralegal or Document Intelligence, or any segregation model. The policy says personal information may be disclosed to processors including marketing and advertising agencies and web analytics companies. For a platform holding client matter documents, that is a real concern, and the policy does not separate account data from uploaded matter content.

A commitment to confidentiality in line with privacy law and with segreto professionale sits on the security page, and the privacy policy states that beyond privacy Lexroom protects professional secrecy and confidential information generally. An operational control is attached: efficiency monitoring data is visible to Lexroom staff only in anonymized form so that confidentiality and professional secrecy are not compromised. Segregation is documented at tenant level. Each customer has a dedicated and exclusive virtual space, private library contents are not shared with other customers, and the privacy policy states it is physically impossible for one customer to reach another's. Google and OpenAI are named and contractually excluded from training on customer prompts. Lexroom itself may use prompts for its own benchmarking and fine tuning after stripping personal and confidential data. Matter level walls inside a single firm are not addressed, although law firms are a named buyer.

Not addressed in any located material. The product generates propositions, advisories and memos from natural language input, which is output shaped like advice, and the user base explicitly extends beyond advocates to founders, tax professionals and compliance teams. That combination is precisely where the question bites: a non lawyer receiving an advisory generated from case law is the unauthorized practice scenario, and nothing published addresses it. No statement that output is not legal advice, no positioning on the supervising advocate's role, and no engagement with Bar Council of India rules on advocate conduct or advertising. The paralegal naming is the closest thing to a position and it is a brand choice rather than a disclosure.

Under Lexroom's privacy policy, outputs are always submitted to a human operator who must review them before use, and the research page frames the product as amplifying professional judgment, with the user rather than an algorithm deciding. Coverage is bounded: the platform is stated throughout as available in Italy, Spain and Germany only. No statement that output is not legal advice is published, and the site publishes no terms of service in which such a statement would normally sit. No bar or professional guidance is engaged: neither Italian Consiglio Nazionale Forense material nor any equivalent in Spain or Germany is named.

Nothing published about how the models are governed, evaluated or monitored. No AI policy, no evaluation methodology, no bias or fairness assessment, no accuracy monitoring, no drift statement, no model card, no named governance body and no external standard such as ISO 42001. The absence is conspicuous against the vendor's own subject matter expertise, since its published blog material analyzes the DPDP Act and DPDP Rules compliance obligations in structured detail for its customers, and none of that analytic rigor is turned on its own system. Also unaddressed: whether a corpus described as machine enhanced introduces systematic error into the archive itself, which for a research product built on a proprietary dataset is the governance question that matters most.

The security FAQ describes Lexroom as fully compliant with the GDPR and the new European AI Act. The privacy policy contains a reasoned analysis of automated processing under Article 22, concluding that outputs are not decisions in the strict sense because a human must review them. A Data Protection Officer is named in full, with chambers, tax code and certified email address, though the role covers data protection rather than model behavior. There is no governance framework, no owner of model behavior, no account of what is tested before release, and no certification such as ISO 42001. Nothing is published about uneven output across matter types, parties or populations.

The vendor addresses stewardship directly, but what it says does not hold together. The home page says customer data is not used for training, and the next sentence says free users can opt out of sharing data for model training; both cannot be complete. The privacy policy supports erasure on request, lists processor categories and says appropriate security measures and generally accepted industry standards apply. No retention period is stated for prompts or output, account data is not distinguished from uploaded matter content, and the training position depends on which sentence a reader stops at. The training signal records the contradiction with the quote. Addressing training at all, and giving free users a control, is more than silence.

Private library files are deleted within 30 days of the end of the relationship, and backup copies on Google Cloud within at most 180 days. Prompts are kept for the duration of the relationship, and customer registry and billing data for ten years under Italian accounting law. Deletion follows the same clock, and data subject requests are answered within 30 days. Data access is limited to a small number of technical staff with special authorizations, with all access and downloads monitored, an enforced password policy, encrypted stored passwords and single sign on over SAML and OAuth 2.0. Google and OpenAI are named as the AI providers, other recipient categories are listed, and a software bill of materials is available on request. A formal emergency procedure provides notification to the customer as soon as possible by certified email or another agreed secure channel. Encryption is AES 256 at rest with a bring your own key option, vulnerabilities are scanned daily, and external penetration tests run with summaries available on request.

Terms exist, were read, and run one way. The terms and conditions make the user indemnify the company and its officers, agents and employees against third party claims, demands, damages, penalties, losses and actions, including reasonable attorneys' fees, and an entire agreement clause incorporates the privacy policy. No vendor commitment was found: no output warranty, accuracy undertaking, service level, liability position, or remedy if a generated citation, advisory or memo is wrong. The record rests on the published terms, which allocate risk in one direction.

Nothing published states who bears the loss when the output is wrong. No terms of service, master agreement or customer contract appears on Lexroom's site. The footer offers only a privacy policy and a cookie policy, and the privacy policy describes the data processing agreement as signed at contract activation rather than published. No indemnity, liability cap, warranty or carve out is stated. Lexroom states that it holds cyber risk insurance with a carrier rated double A by Standard and Poor's, with a copy of the certificate available on request, and that a service level agreement is attached to the corporate plan contract. Neither sets out recourse for a wrong answer; the insurance covers Lexroom's own exposure.

Interoperability is claimed at the right level for the public sector product and documented nowhere. The vendor states that Courtroom and PUBSEC are interoperable with existing and new systems and are transacted with over API, which is the correct shape for selling into court registries, and the January 2026 CADRE ODR partnership is a named integration into an online arbitration workflow. What does not exist: any API documentation, any authentication or scope detail, any statement of what moves in which direction, and any named integration for private practice. No document management system, no practice management platform, and nothing addressing how a firm gets its own matter files in and its work product out.

Data flow through the Microsoft Word add in is set out rather than merely named: the privacy policy states that the add in involves no additional data processing and that prompts formulated through it are directed straight to Lexroom without retention by or access from Microsoft or any other intermediary. Single sign on is specified by protocol, compatible with Google and Microsoft accounts and with SAML and OAuth 2.0. No document management, practice management, email or electronic filing integration is named, no API or developer documentation is published, and no integrations page exists.

One strong deployment statement sits beside silence for everyone else. For the public sector products, the vendor says the agents are owned and hosted by the courts themselves, which is plain customer hosting and fits a judiciary that cannot put case data on a startup's infrastructure. For the commercial product, no hosting provider, region, residency commitment or private or single tenant option is published. Participation in AWS, Microsoft and Google startup programs is not a hosting disclosure. Data residency is a live question in India under the DPDP framework the vendor itself writes about.

All data are processed on servers located exclusively in the European Union, the security page states, with the principal server in the Netherlands and backup redundancy across several EU data centers against extreme weather. The privacy policy adds that Lexroom keeps data long term on servers within the European Union. It discloses the one exception precisely: limited transits to the United States for website hosting only, covered by EU standard contractual clauses and the Data Privacy Framework. The tenancy model is stated, with each customer given a dedicated and exclusive virtual space, and a bring your own key option is offered for encryption. Lexroom operates no data center of its own and relies on cloud services listed in the Italian national cybersecurity agency's digital infrastructure catalog. There is one region and no option outside the EU.

No certification of any kind was found, and the security language cannot be checked. The privacy policy says appropriate security measures are taken against unauthorized access, alteration, modification, disclosure or destruction, and that generally accepted industry standards are followed, without naming a standard, auditor, scope or date. Nothing was found on SOC 2, ISO 27001, ISO 42001, CERT-In empanelment or any Indian assurance framework, and there is no trust center, security page or route to request documents. Court customers host the software themselves, which takes much of this question off their hands, but nothing is published for anyone else.

Certification to ISO 27001:2022 is in place. External penetration testing is stated, with auditors described as external and a summary report available on request. Daily vulnerability scanning is described with findings classified and prioritized, and a software bill of materials and a cyber insurance certificate are each offered on request. No auditor is named, no certificate number, issue date or coverage period is published, and no report is downloadable. A trust center at trust.lexroom.ai is linked from the footer and the security page.

Processor categories are disclosed, a list is promised, and no entity is named. The privacy policy says personal information may go to data processors under the DPDP Act, in categories including hosting providers, cloud computing entities, IT service firms, marketing and advertising agencies and web analytics companies, and says the vendor will try to publish an updated list of key processors. No such list was found. At the model level, the vendor claims proprietary verifier and graph models but never says whether a third party foundation model sits under the agents, which is what a buyer needs to know.

Lexroom's privacy policy states plainly that Lexroom does not develop its own large language model but relies on third party artificial intelligence, and that the service therefore transfers data to its suppliers, in particular Google and OpenAI. It adds a contractual commitment about what those providers may do: customer prompts are expressly excluded from the AI suppliers' training under the terms agreed with them as subprocessors. Where inference runs is addressed at the level of the estate rather than the route, with all processing on EU servers. No model is named, only the companies they come from, and nothing commits to notifying customers when the supply chain or the models change.

A free tier is quantified and paid plans sit behind a login. Free tier limits are ten paralegal sessions, fifteen searches and thirty document intelligence files per month, which shows the unit and shape of the meter. Plan details require signing in, and no price, currency or range for any paid tier was found, so a buyer cannot compare with an existing subscription without an account. The free tier figures come from an independent review platform's assessment dated July 2026, not a vendor pricing page, and that assessment notes pricing is not fully public and may be out of date, so the source basis is recorded as third party estimated.

Pricing starts at 99 euros per month, excluding VAT, on an annual contract. That buys one Lexroom module, one user, up to 200 documents, the Microsoft Word add in, and the possibility of tailored training: a rate, a unit and a term. Above the entry plan, prices are tailored to requirements, so the enterprise range is a sales conversation. Nothing states what implementation or tailored training adds. No module is priced individually, although modules are the unit of purchase. There is no pricing page as such; the figure appears in a demo booking block repeated across the site.

Scale and jurisdiction are stated precisely, and the vendor is open that the scope is one country. The archive holds more than 16 million judgments and statutes, described as India's National Legal Archive and built for Indian law rather than adapted from a general model. Named users include private practice lawyers, law firms, in house teams, judges and registrars in five or more courts, founders, tax professionals and compliance teams, and published topics include SEBI regulations, AML and CFT, the DPDP Act and Rules, the IT Act and Supreme Court authority on Section 65B evidence certificates. Covering one jurisdiction is a limit for buyers outside India, not a disclosure failure. The archive is not broken down: no list of courts, tribunals and forums, no historical depth, and no update lag or refresh schedule, so a practitioner cannot confirm their forum is covered or current.

Lexroom gives seven segments their own page: law firms, companies and public administration on the organization side, and advocates, notaries, accountants and labor consultants on the professional side, which covers private practice, in house and government use explicitly. Jurisdiction is bounded and repeated on every page, with the platform stated as available in Italy, Spain and Germany. Subject coverage is counted rather than listed: more than fifteen subject modules are said to exist, each built and verified individually, and modules are the unit a customer buys, but they are not named. No statement of firm size or of what the product does not support is published.