J
Jusbrasil

Jusbrasil is Brazil's largest legal information platform, operated by Goshme Soluções para a Internet Ltda. of Salvador, Bahia. It indexes and republishes material from official public sources, being official gazettes and court systems, and describes its base as more than 1.2 billion legal documents covering case law, legislation, pleadings and articles, used by millions of people ranging from the general public to advocates. Its AI assistant, Jus IA, launched in March 2025 for legal professionals and shipped with three skills: conversational legal research over the corpus, drafting of pleadings and other legal documents from case details supplied by the user, and analysis of references. The vendor states that alongside established third-party models it has its own foundational model trained on that legal database, developed in partnership with the Brazilian company Maritaca AI, and that the assistant is continuously calibrated by Brazilian legal specialists in a human-in-the-loop process. Two mechanisms distinguish the product from a general chatbot: every reference Jus IA uses carries a direct link into the Jusbrasil base, and at the end of each response the assistant automatically validates the legal and case law references it has just cited, checking whether each is present in the base and flagging any that need attention or could not be found. Users can upload PDF or Word files for summarisation and analysis, organise conversations by client or case number, and pull the full text of a decision into the chat. A separate virtual assistant serves non-lawyer users with general information and routes them to advocates registered on the platform. The terms of use, last modified 1 September 2026 with twelve prior versions published, carry a dedicated Jus IA section that prevails over the rest of the agreement, and state that no AI feature substitutes for a qualified lawyer and that Jusbrasil provides no service reserved to advocates under Article 1 of Law 8.906/1994.

Vendor siteSalvador, Bahia, Brazil
Last verifiedSeptember 7, 2026
Compare with other vendors

Capability grades

All 15 axes, graded from public sources on the date shown. Hover a grade to see what the letter means on that axis.

BB on AI CentralityThe models are the engine of a core capability, layered on a product that would still function without them as a document or workflow system.

AI Centrality

How much of the product is actually AI. Whether the machine learning is the mechanism the buyer is paying for or a feature layered onto conventional software, and whether the vendor is specific about which is which.

Models do substantial work on top of a system of record that would function without them, and the vendor is explicit about which half is which. The model half is real and named: Jus IA is a generative assistant launched in March 2025 with three shipped skills, being conversational legal research, drafting of pleadings and other documents, and reference analysis, all grounded in a corpus the vendor states exceeds 1.2 billion documents. The vendor states it built its own foundational model trained on that corpus in partnership with Maritaca AI, alongside established third-party models, and that the assistant is continuously calibrated by legal specialists in a human-in-the-loop process. A second generative surface serves non-lawyers as a triage chatbot. The other half is not models and long predates them: the platform has indexed official gazettes and court systems for roughly two decades, and the search, case-tracking, legislation, doctrine, pleadings and community publishing products all function without generative capability. Strip the models out and a very large legal database business remains, which is why this is not the top grade. Terms of use of 1 September 2026, product surfaces and launch material read 7 September 2026.

Source: Vendor Published
AA on Citation Accuracy and Hallucination DisclosureMeasured accuracy is published with the test set described and the failure modes named. Output grounds to primary authority the reader can open, citation status is checked, and the system states when it found no support.

Citation Accuracy and Hallucination Disclosure

Whether the vendor publishes measured accuracy on citations and assertions, grounds output to primary sources, and says plainly what its system does when it does not know. Legal has a documented public record of fabricated citations reaching filed briefs, so an untested claim of accuracy is not evidence.

Accuracy is measured, the design of the measurement is published, and the vendor names where it loses. The measurement: a blind comparative evaluation of four systems on real legal tasks, with Jus IA tested against ChatGPT in paid and free versions on GPT-5 and against Gemini 2.5 Pro, using identical neutral instructions. Forty-eight practising Brazilian lawyers scored the outputs without knowing they were assessing AI systems or which system produced which answer; their profile is disclosed as one to ten years in practice, 89.58 per cent holding a postgraduate qualification, drawn from civil, labour, tax, administrative, criminal and commercial practice. Five use intentions define the test set, being document analysis and summary, explaining a legal concept, mapping a jurisprudential position, generating or adjusting a document, and finding precedents. Four scoring criteria are defined, one of which is reliability, expressly measuring whether cited sources exist and support what was asserted and identifying hallucination. Results are published by criterion with means and standard deviations, and the headline is that Jus IA outperformed in 70 per cent of cases. The failure modes are named rather than buried: the vendor publishes that Gemini beats it on linguistic fluency when generating or adjusting documents, attributes that to its own prioritisation of correctness over naturalness, and states that it ties on completeness for document summary. The shipped mechanism sits alongside the measurement: every citation of legislation, case law, súmula or other normative content is automatically checked against the corpus at the end of each response, with references that were not found or need attention flagged in the chat, and every reference carries a link into the source document. Two limits are stated rather than glossed. The headline figure is a relative preference rate against three named generalist systems, not an absolute accuracy or hallucination rate, so a buyer cannot say how often Jus IA is wrong in isolation. And the per-criterion charts are images that did not resolve to machine-readable values on this channel, with the full study behind a link shortener. Study page and product surfaces read in full 7 September 2026.

Source: Vendor Published
BB on Autonomy and Oversight ModelA written commitment that the models work alongside a supervising lawyer, with real review surfaces, short of the full control structure: commonly the threshold at which the system stops or what happens after it is wrong.

Autonomy and Oversight Model

What the system decides on its own, what a lawyer must approve, and whether the vendor documents where the review point sits. A tool that drafts under review and a tool that files without one are different products and different risks.

The oversight position is published, repeated across two separate agreements, and stops short of describing behaviour at the edge. What is published: clause 4.4.2 of the terms of use states that no artificial intelligence feature substitutes the work of a qualified lawyer and that the user is responsible for always reviewing or consulting a qualified professional; clause 5.2.1 states that responses may contain errors or omissions and places verification on the user, expressly including where outputs are used before a public body; clause 5.6.2 states that final responsibility for use of Jus IA and for generated content is exclusively the user's; and the separate enterprise terms for organisations carry a matching obligation at clause 4.3.2 requiring the client to analyse and validate content generated by AI tools. The consumer-facing assistant is separately fenced at clause 6.2.1, which states that its answers are informational and educational only and never constitute legal advice. The vendor also describes continuous human-in-the-loop calibration of the assistant by legal specialists. What is missing is the behavioural half. Nothing published describes an escalation path, a point at which the system declines to proceed on substance rather than on policy grounds, or what a user sees when the model is operating outside the corpus. The product is a chat assistant with no autonomous action surface, so several limbs of this band do not bite and are named rather than penalised: there is no agent acting unattended, no threshold at which the system files, sends or commits anything, and no automated action to define a boundary for. Both agreements and the product surfaces read 7 September 2026.

Source: Vendor Published
BB on Operational and Outcome EvidenceReal deployment evidence with substance, short of full attribution or measurement: a named customer without figures, or figures without the named customer.

Operational and Outcome Evidence

Named, dated evidence that the product works in production at real firms or legal departments. Case studies with figures and identified customers count. Unattributed testimonials and launch announcements do not.

Scale is evidenced in the vendor's own numbers and named customers are absent. What is published: more than 1.5 million searches performed through Jus IA, more than 4 million pleadings written with it, an average satisfaction score of 89 per cent, a corpus of more than 1.2 billion documents, and more than 90 million decisions from 96 tribunals. Three testimonials appear with the full names of practising lawyers rather than initials or job titles alone. The blind study is the strongest operational artefact on the record and is credited on the Citation Accuracy axis rather than counted twice here; what belongs here is its scale, being 48 evaluating lawyers recruited across regions and specialisms. A separate industry survey of 1,500 respondents was run with the São Paulo section of the Brazilian bar, Trybe and ITS Rio, which evidences the vendor's standing rather than its outcomes. What is absent is any named institutional customer, any case study tied to a firm, and any method behind the satisfaction and volume figures. The usage numbers are counts of activity rather than measures of result: nothing states how many of the 4 million pleadings were filed, revised or discarded. Product, pricing and launch surfaces read 7 September 2026.

Source: Vendor Published
BB on Privilege and Confidentiality PostureSubstantive published commitments on confidentiality and training use, short of the full picture: commonly silence on segregation between users or matters, or on what the underlying model provider may retain.

Privilege and Confidentiality Posture

How client confidences are handled: attorney client privilege and work product treatment, segregation of one client matter from another, whether client data trains any model, and what the vendor commits to in writing rather than in marketing.

Substantive published commitments on confidentiality sit against an express contractual disclaimer of responsibility for professional secrecy, and both belong on the record. The commitments are real and specific. Clause 8.3.4 of the privacy policy states that Jus IA operates independently of the platform's other databases and uses a given user's inputs and generated content only within that user's own environment, without providing that user's information to any third party or other user unless the information is publicly available. Clause 8.3 states that Jus IA material is held in a separate database so that it is not used for other purposes and third parties cannot reach it. Clause 5.3.2 states that Jus IA inputs and generated content are treated confidentially and will not be shared with the general public save under a court order. Clause 4.1.5 states that data is not used to train models. The product FAQ adds internal controls preventing personal information entered by one user from being reached by another, and clause 8.1.2 limits employee and third-party access to what is necessary, under express instructions and a contractual confidentiality obligation. Against all of that, clause 5.4.3 of the terms of use states that the user is solely responsible for the information entered into Jus IA, expressly including information protected by professional secrecy, and that Jusbrasil is not responsible for any violation of professional secrecy or any damage arising from inappropriate use. Professional secrecy is therefore addressed in the agreement, which most records on this axis cannot say, and it is addressed by allocating the entire risk to the lawyer. That is why this is not the top grade: the limb this axis exists for is engaged and answered against the buyer. Terms of use and privacy policy read in full 7 September 2026.

Source: Vendor Published
AA on UPL and Professional Responsibility PostureThe vendor states plainly what the product is and is not, who may use it, and how it supports a lawyer’s competence and supervision duties. Jurisdiction limits are named and any consumer facing surface carries a clear disclosure.

UPL and Professional Responsibility Posture

Whether the vendor is clear that it supplies a tool rather than legal advice, who its audience is, and how it addresses unauthorized practice of law, competence and supervision duties, and jurisdiction limits. ABA Formal Opinion 512 is the reference point.

The boundary is drawn expressly, by reference to the governing statute, in both published agreements and on more than one product surface. Clause 4.7 of the terms of use states that Jusbrasil does not under any circumstances provide legal advice, consultancy or any other activity reserved to the lawyer under Article 1 of Law 8.906 of 1994, the Statute of Advocacy and of the Brazilian Bar Association, and that no legal information on the platform is to be understood as legal consultancy or legal services. The enterprise terms for organisations repeat the same statement at clause 3.5. Clause 4.4.2 states that no AI feature substitutes a qualified legal professional. Clause 4.8 states that the platform neither encourages nor performs any form of client or case solicitation, which is the other half of the Brazilian professional-conduct question and is repeated at clause 10.1(iii). The consumer-facing assistant is separately fenced at clause 6.2.1, which states its answers are informational and educational, never constitute an opinion, consultation or legal advice, and do not substitute a lawyer, with clause 6.3.2 disclaiming any involvement in fee negotiation or in the professional relationship formed. User-published content is bound at clause 10.3(iv) to the Statute, to the Bar's Code of Ethics and Discipline and to its official rulings. The mechanism goes beyond text: clause 2.2 requires a valid Bar registration number for users identifying as lawyers, clause 2.2.1 gates certain features behind validation of that registration, and the enterprise terms gate features on the client nominating Bar-registered users. The statute is named by number and article, the reserved activity is named, the solicitation rule is named, the consumer surface is separately handled, and access to professional features is conditioned on verified registration. Both agreements and the product surfaces read 7 September 2026.

Source: Vendor Published
CC on AI Governance and Bias DisclosureResponsible AI principles are published without a mechanism, a testing regime, or anything a buyer could audit.

AI Governance and Bias Disclosure

Published governance over model behaviour: who owns it inside the vendor, what is tested before release, and what is disclosed about disparate output across matter types, parties, or populations.

A practice is described and no governance framework stands behind it. What exists: continuous human-in-the-loop supervision of the assistant by legal specialists, stated repeatedly and in the terms of use at clause 5.2 as a system incremented by specialists across areas of Brazilian law; a statement in the product FAQ that before releasing new features the vendor performs tests and evaluations to ensure a safe, transparent and reliable experience; and internal audits and access controls asserted in the same answer. The published comparative evaluation is a real testing result and is credited on the Citation Accuracy axis rather than counted a second time here. What is absent is everything this axis asks for beyond practice. No governance framework or standard is claimed for the AI, and no ISO 42001 or equivalent is asserted. Nobody is named as accountable for the assistant's behaviour. No bias, fairness or representativeness disclosure exists at any level, which is a live gap on this product specifically: the corpus is built from judicial and official-gazette material that the privacy policy itself acknowledges may contain sensitive personal data revealing religious, political or sexual orientation, philosophical conviction, participation in political or social movements, and health or genetic information, and nothing published addresses whether outputs drawn from that corpus reproduce patterns in it. No model card, evaluation protocol or red-team result is published. Terms of use, privacy policy and product surfaces read 7 September 2026.

Source: Vendor Published
BB on AI Safety and Data StewardshipSubstantive published policy covering most of the ground, short of the full set: commonly no named subprocessor list or no stated incident practice.

AI Safety and Data Stewardship

Retention, deletion, access control, and what happens to prompts and documents after they are processed. Whether the vendor states its subprocessors and its incident practice, or leaves the buyer to assume.

Substantive published policy across most of the ground, with the two limbs that would lift it absent. What is published: a corporate information security policy and integrated management document for the information security management system, version 1.0 dated 20 January 2026, issued by the vendor's GRC and Security function, published for external use with a downloadable PDF and a version history; that document commits to risk identification, analysis and treatment, to incident management and continuity processes covering prevention, detection, response and recovery, to vulnerability treatment prioritised by impact, to monitoring, traceability and audit of information assets, to awareness training, and to senior management accountability, all structured on a plan-do-check-act cycle. On the data side the privacy policy commits at clause 8.1.2 to limiting employee and third-party access to what is necessary under express instruction and contractual confidentiality; at clause 8.3 to holding Jus IA inputs and generated content in a separate database; at clause 8.3.1 to discarding voice biometric characteristics after transcription and retaining only the text; at clause 8.3.3 to deleting Jus IA data on account deletion; and at clause 8.3.5 to definitive deletion of a conversation and its contents from the platform and internal databases when the user deletes it. Cross-border transfer is disclosed at clause 9.1. What is missing: no subprocessor or cloud provider is named anywhere, so a buyer cannot say who holds the data; and no breach notification commitment to customers is published, the incident process being described as an internal capability rather than a notification duty. Encryption is not specified. The dedicated security page at the main domain returned bot detection on this channel and is the named rebuttal route. Security policy and privacy policy read in full 7 September 2026.

Source: Vendor Published
CC on AI Liability and RecourseLiability is addressed only through a standard limitation clause that disclaims the exposure the product creates.

AI Liability and Recourse

What the vendor stands behind contractually when its output is wrong. Indemnities, caps, carve outs, insurance, and whether any of it is published or only reachable through a negotiated agreement.

The allocation is published in full, readable before signing, and runs almost entirely in one direction. Clause 4.2 puts the platform and services on an as-is footing and reserves the right to withdraw or change them unilaterally. Clause 5.6.2 states that no guarantee of any kind is given as to the results of applying information from the platform in legal practice. Clause 10.2 sets out seventeen enumerated heads of non-liability, including loss of procedural deadlines, errors in information supplied by the courts, discrepancies with official bodies, the user's use of information from the platform, and at sub-clause (xiii) all direct and indirect, incidental, special, punitive and consequential loss including lost profits and moral damages. Clause 10.3(iii) runs the indemnity the other way: the user indemnifies Jusbrasil, its affiliates, directors and employees for any loss arising from use of the platform. Clause 5.4.3 disclaims responsibility for any violation of professional secrecy. No indemnity is given to the customer for anything, including third-party intellectual property claims, no liability cap or floor is stated, no warranty is offered, no service level is committed and no insurance position is published. The enterprise terms repeat the exclusions with no cap either. What keeps this off the floor is that the allocation is comprehensively published and readable rather than absent, that a seven-day refund right is stated at clause 3.4.1, and that the vendor publishes twelve prior versions of the agreement so a buyer can see how the allocation has moved. On the question this axis asks, the published answer is that the lawyer bears the loss. Terms of use read in full 7 September 2026.

Source: Vendor Published
CC on Practice Systems Integration DepthIntegrations are listed as logos or marked as coming, with no documentation an implementer could use.

Practice Systems Integration Depth

How deeply the product reaches into the systems legal work already lives in: document management such as iManage and NetDocuments, Word and Outlook, contract lifecycle management, matter management, e-billing, and court filing systems.

A developer surface exists and is documented, and no legal practice system is named. What is established from first-party material: the enterprise arm sells its solutions in API form, described in its terms at clause 1.1 as available by API, interface or other means; a documentation link is published in the enterprise site navigation and footer; a separate API product line is published; and the consumer terms at clause 2.5.2 treat unauthorised API access as illegitimate, which confirms that authorised API access exists. Two lighter integration surfaces are described in the privacy policy and product pages: browser extensions the vendor publishes, and a WhatsApp sharing path that lets a user send files and audio into Jus IA from the messaging app. What is absent is depth into the systems this axis asks about. No document management system, case management platform, word processor, email client or e-signature product is named anywhere on the surfaces read; no integrations page, connector list or partner directory was located; and nothing describes what moves in which direction. Recorded so the grade is read correctly: the API documentation is published at a URL located in the vendor's own navigation but was not opened on this channel, so its existence is established and its depth is not, and it is the named rebuttal route on this row. Both agreements, the enterprise site navigation and the product surfaces checked 7 September 2026.

Source: Vendor Published
BB on Deployment Model and Data ResidencyDeployment model is stated clearly with partial residency detail, or residency is offered without the processing location being addressed.

Deployment Model and Data Residency

Where the software runs and where the data sits. Multi tenant cloud, single tenant, private deployment, on premises, and whether region of residence is a published option or an enterprise conversation.

Both limbs are addressed and neither is elaborated. On residency, clause 9.1 of the privacy policy states that personal data of users and of people named in proceedings may be stored on servers outside Brazil, expressly including in the United States, because of cloud storage contracted with companies operating data centres there, with clause 9.2 committing to appropriate contractual arrangements. That is a residency disclosure and, for a Brazilian buyer handling Brazilian matter material, an adverse one stated plainly rather than buried. On tenancy, clause 8.3 states that Jus IA information is held in a separate database from the platform's other bases, and clause 8.3.4 states that the system uses a given user's inputs and generated content only within that user's environment. Clause 4.1.5.2 adds that public data and user-entered data are not combined into a single base. Together those describe how one customer's material is separated from another's and from the public corpus. What is absent: no region selection is offered, no Brazilian-only hosting option is published despite the product being sold exclusively into Brazil, no single-tenant or self-hosted arrangement is described, no cloud provider is named, and nothing distinguishes where processing happens from where data is stored beyond the transfer statement. Privacy policy read in full 7 September 2026.

Source: Vendor Published
CC on Security Certifications and Trust CenterBadges appear on the site with no scope, no date, and no report available.

Security Certifications and Trust Center

Independent attestation a buyer can pull without a sales call: SOC 2, ISO 27001, penetration test summaries, a trust center with current reports and named scope rather than a badge image.

A real security governance document is published and no certification is held or claimed. What was located and read: a Política de Gestão Integrada for the information security management system, published for external use on the enterprise site with a downloadable PDF, carrying a version number, a creation date of 20 January 2026 and a named owning function of GRC and Security. It states that the management system is built to the requirements of ISO/IEC 27001 and sets out corporate security directives covering confidentiality, integrity and availability, protection against unauthorised access, use of assets only for authorised purposes subject to monitoring and audit, incident management and continuity, vulnerability and risk treatment, awareness training and senior management accountability. It applies to employees, interns and third parties. What it does not do is claim the certification. Its own opening states that certification of the management system to ISO/IEC 27001 is a strategic objective of the security function requiring a set of requirements to be met, which describes an aspiration and a programme, not an attestation held. Nothing else was located: no certificate, no auditor, no SOC report, no audit period, no scope statement, no penetration test result, no trust centre and no self-serve or gated artefact repository. This is therefore alignment with a named standard, documented at policy level in the vendor's own words, and independently attested nowhere. The dedicated security page on the main domain returned bot detection on this channel and is the named rebuttal route. Security policy read in full and enterprise site navigation checked 7 September 2026.

Source: Vendor Published
BB on Model Supply Chain DisclosureThe supply chain is partly disclosed: providers named without change notification, or architecture described without the providers.

Model Supply Chain Disclosure

Which models sit underneath, whose they are, where they run, and whether the vendor commits to telling customers when that changes. A legal buyer inherits every dependency it cannot see.

The supply chain is partly disclosed, and the disclosed part is the vendor's own model rather than the bought-in ones, which is the reverse of the usual shape. What is named: the vendor states in its own launch material that alongside established third-party models it has its own foundational model trained on its legal corpus, developed in partnership with Maritaca AI, a Brazilian company specialising in adapting language models to particular domains and languages. Trade press quoting the vendor names the partner model as Sabiá and describes the product as combining general-purpose models with it. That satisfies both limbs for one supplier: a provider is identified and its model is named. What is not named is everything else. The privacy policy at clause 4.1.5 refers to the third-party natural language models that feed Jus IA without naming one; the terms of use at clause 5.4.1 refer to partner companies providing natural language processing services without naming one; and clause 8.4 states that AI functionality may depend on third-party technology and models subject to their policies and availability. So a buyer can establish who built the specialised model and cannot establish whose general-purpose models process their inputs. No commitment is given to notify customers when the arrangement changes, and no subprocessor list exists to cross-check against. Terms of use, privacy policy and vendor launch material read 7 September 2026.

Source: Vendor Published
AA on Commercial TransparencyA buyer can learn what this costs without entering a sales process: published rates, the unit being charged, and what implementation adds.

Commercial Transparency

Whether a buyer can learn what this costs without entering a sales process: published rates, the unit being charged, what sits behind an enterprise tier, and what implementation adds.

Published pricing with the unit of charge, the figures, the limits and the terms, all reachable without a sales conversation. Three tiers are named with regular monthly prices: Essencial at R$78.90, Profissional at R$108.90 and Premium at R$208.90, each shown against a promotional first-month rate, with a monthly and annual toggle and a stated annual saving of up to 33 per cent. The unit of charge is published and is the AI message: 30 per month, 150 per month and unlimited across the three tiers. A comparison table quantifies seven further limits per tier, being document uploads per month at 10, 50 and unlimited, documents per conversation at 1, 5 and unlimited, file size at 10MB, 50MB and unlimited, cases with memory at 10, 50 and unlimited, and doctrine access at 3 per month then unlimited, alongside which tiers carry citation validation, procedural insights and notifications, curated news and priority support. The commercial terms are in the agreement rather than only on the page: automatic renewal at clause 3.4, self-serve cancellation at any time with benefits running to the end of the paid cycle, a seven-day refund window at clause 3.4.1, and a statement at clause 3.4.2 that where no minimum period is set there is no lock-in. An enterprise route for firms, companies and public institutions is named with payment by boleto. One discrepancy is recorded because a reader who finds it unaided should find it here too: the page's own title states a first month from R$9.90 while the body shows R$1.90 for all three tiers. Pricing page and terms of use read 7 September 2026.

Source: Vendor Published
BB on Firm and Practice CoverageSegment and practice coverage is described with substance, short of the boundaries: what is supported is clear, what is not is left open.

Firm and Practice Coverage

Who the product is actually built for. AmLaw, midlaw, small firm and solo, in house departments, government and courts, and which practice areas are supported rather than merely claimed.

Coverage is described in real numbers across a single jurisdiction, and the boundaries of the offer are left unstated. What is published: more than 90 million decisions from 96 tribunals, a corpus stated at more than 1.2 billion documents, daily updating with new normative acts and recent decisions from courts across the country, and separate collections for case law, legislation, doctrine at more than 1,900 works, real filed pleadings, and community-contributed models. Practice areas are addressed as the whole of Brazilian law rather than a named list, with the study's evaluator panel drawn from civil, labour, tax, administrative, criminal and commercial practice. Buyer types are named across two arms: individual lawyers and law students on the consumer platform, and law firms, corporate legal departments, companies and public institutions through the enterprise arm, which publishes separate surfaces for each. Access to professional features is conditioned on Bar registration. What is absent is any statement of limits: no depth or lag is given per tribunal, nothing states which courts are covered in full text against summary only, no practice area or matter type is named as out of scope, and the jurisdictional boundary is implied by the sources rather than stated. Product, pricing and enterprise surfaces read 7 September 2026.

Source: Vendor Published
Pricing

From R$78.90 per month

  • Jus IA is sold as three individual subscription tiers with published prices, from R$78.90 a month at the entry level to R$208.90 for the unlimited tier, with a discount of up to 33 per cent for paying annually. What separates the tiers is how much you can use rather than which product you get: monthly AI messages run from 30 to 150 to unlimited, with matching limits on how many documents you can upload, how large they can be, how many you can put in a single conversation, and how many cases you can keep with saved history. Access to the case law, pleadings and templates collections is included at every level, and doctrine is capped at three works a month on the entry tier and unlimited above it. Automatic citation checking is included on all three. Billing renews automatically, you can cancel yourself at any time and keep the service until the end of the period you have paid for, and there is a full refund if you cancel within seven days. Firms, companies and public bodies are quoted separately.

Three published subscription tiers, charged per individual user and billed monthly or annually, with a stated annual saving of up to 33 per cent. Regular monthly prices are R$78.90 for Essencial, R$108.90 for Profissional and R$208.90 for Premium, each currently shown against a promotional first-month rate. The figure recorded here is the Brazilian real amount as published, not a converted United States dollar figure; no conversion is made because no sourced rate exists and the published price is in reais. The unit of charge is the AI message: 30 per month on Essencial, 150 on Profissional and unlimited on Premium, the unlimited tier stated to be subject to abuse-prevention guidelines. Seven further limits are quantified per tier, being document uploads per month at 10, 50 and unlimited, documents per conversation at 1, 5 and unlimited, maximum file size at 10MB, 50MB and unlimited, cases with memory at 10, 50 and unlimited, and doctrine access at 3 works per month then unlimited. Citation validation is included on every tier; procedural insights and notifications, curated news, priority support and early access to releases are tier-dependent. Every tier includes access to the Jusbrasil corpus of case law, models and pleadings. Commercial terms come from the agreement rather than the page: automatic renewal at clause 3.4 of the terms of use, self-serve cancellation at any time with benefits running to the end of the paid cycle, refund available within seven days of subscribing at clause 3.4.1, and a statement at clause 3.4.2 that absent a stated minimum period there is no lock-in. A separate route for law firms, companies and public institutions is published, priced on application and payable by boleto, with multi-user accounts available by changing plan in the platform or through the sales team. One discrepancy on the vendor's own page is recorded: its title states a first month from R$9.90 while the body shows R$1.90 across all three tiers. Pricing page and terms of use read 7 September 2026.

Legal Signals

What each signal means

A signal records what public sources say on the date shown. It is not a grade and it is not a recommendation. Where a signal reads Not addressed, it means the index did not locate the material in public sources on that date, which is a statement about disclosure rather than about the product.

Confidentiality and Privilege

Client Data in Training

Can material a lawyer puts into this product be used to train a model?

Never, in the contract

The published terms prohibit training on customer content. Not a policy page, the agreement.

The prohibition sits in the agreement rather than on a policy page, and the reason is structural: clause 1.3.1 of the privacy policy states that the policy is itself part of the terms of use, and clause 1.2 of the terms makes acceptance of it a condition of use, so its clause 4.1.5 is a contractual term. That clause states that data is not used for the training of proprietary models, giving as its example the third-party natural language models that feed Jus IA, but is used for improvement of the service as a whole. Three qualifications travel with it and none is minor. First, the scope of the word proprietary: the example given is third-party models, while the vendor separately states it has built its own foundational model on its corpus with Maritaca AI, and nothing published says whether the prohibition reaches that model. Second, clause 5.4.2 of the terms grants a broad right over the same material, permitting information shared by the user to be used for research, service quality analysis, and the development and improvement of Jus IA; read with clause 4.1.5 the two are coherent, the improvement right being granted and model training carved out of it, but the improvement right is wide and is not limited to de-identified or aggregated content. Third, clause 4.1.5.2 states that inputs and outputs may be accessed by Jusbrasil to validate the quality of generated content and improve response accuracy, which is human or systematic access to matter material for product purposes. The lower values were tested before this one was taken: the policy-only value's own words require that no matching term be located in the published agreement, which is false here, and the permissive values assert a reserved right to train, which the agreement denies. Terms of use of 1 September 2026 and privacy policy of 22 June 2026 read in full 7 September 2026.

Source: Vendor PublishedOs dados não são utilizados para o treinamento de modelos proprietários (por ex.: os modelos de linguagem natural de terceiros que alimentam o Jus IA), mas são usados para o aprimoramento do serviço como um todo.As of Sep 7, 2026Evidence

Prompt and Output Retention

How long does the product keep what a lawyer typed, and can that be set to zero?

Customer set, zero available

The customer sets the retention window and no retention is an available setting.

The customer controls retention and the control reaches zero, exercised item by item rather than as a policy setting, and the limitation is stated here because it matters. What the customer can do: clause 8.3.5 of the privacy policy states that the user may delete conversations and the personal data contained in them, including data in inputs, outputs and attachments, directly in the Jus IA chat interface, and that because the action is voluntary the information for that interaction is deleted from the platform and from the company's internal databases. Clause 8.3.5.1 extends the same effect to deleting a grouped case, and clause 8.3.3 states that deleting the account deletes information collected through Jus IA. That is definitive deletion at the user's initiative, reaching zero, described as reaching the internal bases rather than only the visible history. What the customer cannot do is set a period. Absent deletion, clause 8.3 states only that inputs, documents, generated content, feedback and generated pleadings are held for as long as necessary for provision of the service, internal analysis and any later legal obligations, with no number attached, so the default is disclosed without a period. One narrower commitment is firmer: clause 8.3.1 states that direct voice recordings are held only temporarily for transcription, after which voice biometric characteristics are discarded and only the text remains. Privacy policy read in full 7 September 2026.

Source: Vendor Publishedas informações daquela interação específica, incluindo dados pessoais e demais conteúdos, são excluídas da Plataforma e das bases de dados internas da companhiaAs of Sep 7, 2026Evidence

Ethical Walls and Matter Segregation

Does retrieval respect the firm’s ethical walls, or can the model read across them?

Own model, documented

The product maintains its own permission model, documented, requiring the firm to keep it aligned.

The product maintains its own segregation model, documented, and it operates at the user rather than the matter or firm level. What is documented: clause 8.3.4 of the privacy policy states that Jus IA operates independently of the platform's other databases and uses a given user's inputs and generated content only within that user's environment, without supplying that user's information to a third party or another user unless the information is publicly available. Clause 8.3 places Jus IA material in a separate database so third parties cannot reach it. The product FAQ states that internal controls prevent personal information entered by one user from being accessed by another. A matter-shaped boundary exists inside the account: workspaces group conversations by theme, client or case number, and each new conversation in a workspace draws on that workspace's history, so context is scoped to a matter rather than pooled across the account. What is absent is the firm-level question this signal asks. Nothing inherits an access model from a document or case management system, because no such integration is named. Nothing addresses walls between users inside one organisation: the consumer terms treat the account as individual and non-transferable at clause 2.6, multi-user plans give each login its own access without describing what one login can see of another's work, and the enterprise terms leave user nomination and revocation to the client without describing separation between them. A firm buying seats would have to maintain any wall itself. Privacy policy, both agreements and the product surfaces read 7 September 2026.

Source: Vendor PublishedO sistema utiliza exclusivamente os Inputs e Conteúdo Gerado por determinada Pessoa Usuária apenas dentro do ambiente daquela Pessoa UsuáriaAs of Sep 7, 2026Evidence

Third Party Request and Subpoena Notice

If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?

Disclosure addressed, notice absent

Published terms or policy address disclosure to authorities or in response to legal process, and no commitment or reservation regarding customer notice is located anywhere. The vendor has told the customer that data can leave and has said nothing about whether the customer hears of it.

Compelled disclosure is addressed in three separate clauses and customer notice appears in none of them. Clause 5.2.2 of the privacy policy reserves the right to share any personal data the vendor believes necessary to comply with a legal obligation or the orders of competent authorities, or to protect its own rights, expressly including responding to judicial requests and sharing with governmental and judicial authorities. Clause 4.1.13 restates the position for judicial orders, search warrants and subpoenas where the vendor believes in good faith that it is necessary. Clause 5.3.2 addresses the Jus IA material specifically and is the sharpest of the three: inputs and generated content are treated confidentially and will not be shared with the general public save in the case of a court order determining that they be shared. So the vendor has considered what happens to a lawyer's prompts and drafts under compulsion and has published the answer. What it has not published is any commitment to tell the customer, any reservation of discretion over notice, any undertaking to seek a protective order or to narrow production to what is legally required, or any transparency report. The threshold is also broad on its own terms, extending to what the vendor believes necessary to protect its own rights rather than only to what is legally compelled. No law enforcement guidelines page was located on either domain. Privacy policy read in full and both agreements checked 7 September 2026.

Source: Vendor PublishedReservamo-nos o direito de compartilhar quaisquer dados pessoais que acreditamos serem necessários para cumprir uma obrigação legal ou ordens de autoridades competentes ou, ainda, para proteger os nossos direitos.As of Sep 7, 2026Evidence
Accuracy and Authority

Primary Law Corpus Provenance

Where does the law in this product come from, and does the vendor have the right to use it?

Sources named and licensed

The vendor names its primary law sources and the licence or public domain basis for each, with an update cadence.

The primary law sources are named individually, the access basis is stated, and the update cadence is published. The enterprise terms carry a formal definition of Fontes Originais as publications in the official gazettes and the websites of state and federal courts made electronically by different organs of public administration, plus data from other public-power sources such as the federal revenue service, collected through partner suppliers. The consumer terms at clause 4.10 say the same in narrative form, and clause 4.6 characterises the platform as a locator and indexer that exercises no editorial control over reproduced content, with clause 4.6.1 limiting its own changes to formatting, language simplification and standardisation. The basis is stated rather than assumed: clause 2.3.4.1 of the privacy policy identifies the material as documents publicly available for consultation in the official gazettes and court systems. Cadence is published as daily updating with new normative acts and recent decisions. Rights in the compilation are asserted rather than left open, the enterprise terms defining the database as one over which Jusbrasil Soluções holds sole and exclusive intellectual property rights. Two gaps are named. No licence basis is stated for the doctrine collection of more than 1,900 works, which are copyrighted publications rather than public law. And community-contributed pleadings and models are not licensed but assigned: clause 9.4 of the consumer terms takes an irrevocable, perpetual, worldwide, unremunerated assignment of the economic intellectual property rights in user-submitted material. One limit of this session is recorded rather than presented as an absence: whether Jusbrasil's corpus or its republication of judicial material has been adjudicated in Brazil was not searched, and this row rests on what the vendor publishes. Both agreements and the privacy policy read 7 September 2026.

Source: Vendor Publishedpublicações nos Diários Oficiais e sites dos Tribunais, Estaduais ou Federais, realizadas eletronicamente por diferentes órgãos da administração pública, e dados de outras fontes do Poder Público, como a Receita FederalAs of Sep 7, 2026Evidence

Good Law Verification

Does the product tell you when the authority it just cited has been overruled?

Not addressed

No located public material addresses whether authority is checked for subsequent history.

No located public material addresses whether authority is checked for subsequent history, and the mechanism that looks closest answers a different question. What the product does is verify existence: at the end of each response every citation of legislation, case law, súmula, portaria or other normative content is automatically checked against the Jusbrasil base, and references that are not found or need attention are flagged in the chat, with links into each source document. That is a check on whether the authority is real, which is credited on the Citation Accuracy axis and is not counted again here. It is not a check on whether the authority still stands. Nothing published states whether a decision has been reversed, vacated, distinguished or superseded, whether a súmula has been revised or cancelled, or whether a cited statutory provision remains in force, and no treatment flag, currency indicator or licensed citator is described. The corpus is stated to be updated daily with recent decisions, which keeps the collection current without telling a user that the particular judgment in front of them has been overtaken. The analogue closest to a treatment signal in Brazilian practice, whether a cited position reflects the prevailing understanding of the court, is described only as a research technique in platform content rather than as a product capability. Product surfaces, support material and both agreements checked 7 September 2026.

Source: Operator VerifiedAs of Sep 7, 2026Evidence

Refusal and Uncertainty Behaviour

What does the product do when the answer is not in the corpus?

Documented

The vendor describes refusal or abstention behaviour in public materials.

Refusal behaviour is documented, and it is documented in the agreement rather than in a help page, which is unusual on this signal. Clause 5.2.3 of the terms of use reserves the right to interrupt use or decline to answer questions from anyone using Jus IA to deviate from the product's stated purposes, which are named as legal research, document creation and analysis of documents and references. Clause 5.2.2 records a second and different ground: because third-party technologies are involved and their policies and codes of conduct may apply, the vendor reserves the right not to answer prompts that potentially conflict with those third-party policies or with the terms, which tells a reader that upstream model policies shape refusals. Clause 5.6.4 enumerates what the user may not attempt, expressly naming prompt injection, prompt leaking, jailbreaking and data extraction outside the product's purposes, alongside categories of content the system will not produce. So the grounds for refusal are published, specific and contractual. What is absent is the uncertainty half. Nothing describes what the assistant does when it cannot ground an answer in the corpus, whether it declines rather than generating, or whether any confidence or coverage signal is exposed. The citation validation that flags references it could not find is the nearest thing to such a signal and is credited on the Citation Accuracy axis rather than counted twice here. No demonstration, evaluation or published result of refusal behaviour exists, which is what separates this from the top value. Terms of use read in full 7 September 2026.

Source: Vendor PublishedO Jusbrasil se resguarda ao direito de interromper o uso ou não responder questões e/ou demandas de qualquer pessoa que utilize Jus IA com o objetivo de desviar das finalidades do produtoAs of Sep 7, 2026Evidence

Fabricated Citation Record

Does a public court record exist involving output from this product?

None located

No court order, opinion or disciplinary record naming this product has been located as of the date shown. This is a statement about the public record, not a finding about the product.

No court order, opinion or disciplinary record naming Jusbrasil or Jus IA as the product behind a fabricated citation was located as of 7 September 2026. Searches were run in Portuguese on the company name, the product name and the terms Brazilian courts use for the conduct, alongside the AI hallucination cases database maintained by Damien Charlotin. Brazilian decisions sanctioning lawyers for invented case law and doctrine were located and none names this product; the Santa Catarina matter that appears alongside Jusbrasil in coverage does so because the vendor's own launch material cites it as the risk its citation validation is designed to prevent, which is the opposite of a finding against the product. This is a statement about the public record and not a finding about the product. One feature of this record bears on how the statement should be read: the vendor publishes a citation validation step that checks each reference against its corpus and flags what it cannot find, and a blind evaluation whose reliability criterion expressly measures hallucination, so the vendor has instrumented the question rather than left it to the record.

Source: Operator VerifiedAs of Sep 7, 2026Evidence
Professional Responsibility

Bar Guidance Alignment

Has the vendor engaged in public with the ethics opinions its buyers are bound by?

Generic reference

Public materials refer to professional responsibility in general terms without naming guidance.

Professional responsibility is engaged in general terms and no bar guidance on the use of AI is named. What exists on the AI question is substantive but generic: clause 4.4.2 of the terms states that no AI feature substitutes the work of a qualified lawyer and that the user must review or consult a qualified professional; clause 5.2.1 states that responses may contain errors and places verification on the user, expressly including where outputs are used before a public body; clause 6.2.1 fences the consumer assistant as informational only; and the enterprise terms at clause 4.3.2 make validation of AI-generated content a client obligation. None of that names an ethics opinion, a bar resolution or a regulator's guidance on lawyers' use of AI. Named bar authority does appear on this record, and it is recorded here so a reader sees why it did not lift the value: clause 10.3(i) binds user-published articles to Law 8.906 of 1994 and to the Bar's rules, clause 10.3(iv) names the Bar's Code of Ethics and Discipline and its official rulings, and clause 4.8 addresses solicitation of clients. Those are named instruments addressing advertising and solicitation conduct, not AI use, and this signal measures engagement with AI guidance. The vendor's standing with the profession is closer than most, having co-run a 1,500-respondent survey on AI in law with the São Paulo section of the Bar, but a joint research exercise is not guidance mapped to the product. Both agreements and the product surfaces read 7 September 2026.

Source: Vendor PublishedNenhuma funcionalidade baseada em inteligência artificial substitui o trabalho humano de um advogado e/ou profissional do direito habilitado.As of Sep 7, 2026Evidence

Billing and Fee Posture

Does the vendor address what happens to the bill when the work takes an hour instead of six?

Savings claims only

Public materials claim time savings without addressing billing or disclosure, and the product sits inside a fee relationship between a lawyer and a client where those savings would change the bill.

Time savings are marketed to the lawyer and nothing addresses what happens to the client's bill. The product sits squarely inside a fee relationship: it is sold to advogados and to law firms, and its outputs are the research and the pleadings those lawyers bill for. The savings claims are explicit and are aimed at compressing exactly the work that is billed, the product pages promising legal research in minutes, hours of manual searching saved, less time in research, and reduced time to produce pleadings, with more than 4 million pleadings written through the assistant. Against that, no published material addresses billing, fee treatment or client disclosure. Nothing tells a firm how to treat an hour that became ten minutes, nothing offers a per-matter record of AI-assisted work that could support a fee narrative, and no guidance on disclosing AI assistance to a client was located on any surface. The agreements are silent on the question: they address what the user pays Jusbrasil, at clause 3 of the terms of use, and never what the user's client pays the user. The absence is worth naming precisely because the vendor engages professional responsibility carefully elsewhere on this record, drawing the unauthorised-practice boundary by statute and article and gating features on Bar registration, so the omission is not a general silence about professional obligations. Product surfaces, pricing page and both agreements checked 7 September 2026.

Source: Vendor PublishedAs of Sep 7, 2026Evidence

Outside Counsel Guideline Readiness

Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?

On request only

The material exists behind a sales conversation or an executed agreement.

The forwardable material exists behind an executed agreement rather than on a page. The enterprise terms at clause 2.1.2 record that personal data is processed in accordance with a Termo de Tratamento de Dados executed with the client, which is a data processing agreement negotiated per customer and not published; clause 1.2 names the Contrato de Prestação de Serviços Jusbrasil Soluções and clause 1.2.1 the Contrato para Organizações as the instruments the client signs. A firm can therefore obtain a processing agreement, and cannot obtain one before entering a commercial conversation. What is published falls short of the pack this signal describes. No subprocessor list exists on either domain: the privacy policy describes categories of suppliers and partners and names none of them, and no cloud provider is identified anywhere despite clause 9.1 disclosing storage outside Brazil. On the model side the disclosure is partial rather than absent, the vendor naming Maritaca AI as the partner behind its own foundational model while referring to the third-party natural language models that feed Jus IA without naming one, so a firm can answer half of a client's question about whose model sees its content. Two artefacts that would bear on this were located and not read on this channel and are the named rebuttal route: the enterprise Aviso de Privacidade and the Guia de Boas Práticas, both published in the enterprise site footer. Both agreements and the privacy policy read 7 September 2026.

Source: Vendor Publishedinformações pessoais (por exemplo: email corporativo) que serão tratadas em conformidade com o Termo de Tratamento de Dados celebrado com o ClienteAs of Sep 7, 2026Evidence

Court Disclosure Support

If a judge’s standing order requires an AI disclosure, can the product produce one?

Not addressed

No located public material addresses court disclosure or verification certification.

No located public material offers a record of AI-assisted work that a lawyer could put in front of a court, and the vendor has considered the court-facing scenario without providing one. Clause 5.2.1 of the terms of use expressly contemplates it, stating that the user is entirely responsible for the use made of Jus IA including where outputs are used before a public body. That allocates the risk and stops there. What the product retains is not a disclosure record: conversations are stored and can be grouped by client or case number, and clause 5.4 records that questions, uploaded documents and chat history are kept, but none of it is offered as an exportable account of what the AI did. Nothing records which model produced a given output, what it retrieved, or what a person verified before the result was used, and no certification, template or standing-order guidance for a filer is published. The citation validation panel, which shows which references were confirmed against the corpus and which were flagged, is the closest artefact and is credited on the Citation Accuracy axis rather than counted twice here; it is in any event a check on the answer rather than a record of the process, and it is not described as exportable. Product surfaces, support material and both agreements checked 7 September 2026.

Source: Operator VerifiedAs of Sep 7, 2026Evidence
Contact

Correct a record, or ask how something was graded

Every grade and every signal on this index is drawn from public sources and dated. If a record is wrong, out of date, or missing an artifact the index did not locate, send the source and it will be reviewed and the record redated. Vendors are welcome to submit documentation. Nothing on this index is for sale, including a listing, a placement, or a grade.

AI Legal Index

The AI Legal Index is an independent index that tracks changes to AI vendors in legal. It holds 61 vendors across 9 categories, each graded on the same 15 capability axes and recorded against 12 legal signals, from public sources. No vendor pays for inclusion, placement, or rating.

Index Status
Last index update
September 7, 2026
The AI Legal Index is an editorial reference. It is not a regulatory body, not a law firm, and nothing published here is legal advice or a recommendation to retain or avoid a vendor. Records are verified against published sources, bar guidance and public court records. Where a record reads not addressed, the material was not located in public sources on the date shown. See the Methodology page for evaluation standards and limitations.
© 2026 AI Legal Index
3801 N Capital of Texas Hwy, Ste E240 · Austin, TX 78746