T
Turivius
Turivius is a Brazilian legal intelligence platform combining case law research, jurimetrics and generative AI over a database the company reports at more than 130 million judicial decisions drawn from over 100 Brazilian courts, updated daily. It is sold in two environments with separate logins: Turivius Clássica, the established product covering jurisprudence search, jurimetrics and knowledge management, and GPTuri+, the newer AI environment.
GPTuri, launched in 2025, is a generative legal assistant that answers in natural language or by voice, builds the search itself from a description of the case rather than requiring Boolean syntax, and produces finished work product: executive summaries, comparative tables, draft pleading structures, opinions, risk maps, counterparty analyses and reports. Every answer cites the decision it rests on so the user can open the original before relying on it, and documents the user uploads, such as contracts, pleadings and case files, can be questioned directly with answers referenced to the exact passage.
Unusually, the buyer chooses which underlying model runs each task, with GPT, Claude and Gemini all available in the same interface and all reading the same corpus, and a web search layer brings legislation, binding summaries and doctrine into context. The Jurimetria Agent, which the company describes as the first of its kind in Brazil, takes a legal thesis, searches at scale, analyses up to 500 selected decisions and returns structured tables, trend charts, divergences between courts and outcome predictions.
Projects hold the files, saved decisions and analyses for a case or client and carry that material into new conversations. Customers named on the company's own surfaces include Deloitte, Nestlé, Gerdau, Suzano, BASF, Unimed, BTG Pactual and the law firms Tozzini Freire and Machado Meyer, with published case studies covering Nestlé, Cosan, Suzano and firms including Velloza, Bichara, KLA, FNCA, Bergamini and Benvenutti.
Turivius Portais de Conteúdo Ltda is based in Florianópolis, Santa Catarina. It was founded in 2019 by Danilo Limoeiro, who remains chief executive, and Guilherme Kenzo, out of research work at MIT and the University of São Paulo; the name joins Alan Turing to Gnaeus Flavius, the Roman scribe who wrote down court procedure that had previously been held orally. The company raised an institutional round in 2021 from ABSeed, Domo Invest, Ace Startups, SDx Fund and Allievo Capital, won Nestlé's global innovation prize in 2022, and in 2023 Banco BTG Pactual became a shareholder.
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.
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.
The models are the engine of a core capability layered on a product that functions without them, which is the B band, and this vendor draws the line itself more clearly than most because it sells the two halves as separate environments. Turivius Clássica is the established product covering jurisprudence search, jurimetrics and knowledge management over a corpus the company reports at more than 130 million decisions from over 100 Brazilian courts.
GPTuri+ is the AI environment, launched in 2025, with its own login and its own subdomain, and the company states plainly that the two accesses are independent and that a user signs in to whichever environment they hold. So a buyer can hold the platform without holding the AI, which settles the axis. What sits on the AI side is substantial and named: a generative assistant that builds the search from a natural-language or spoken description of the case rather than requiring Boolean syntax, reads uploaded contracts and pleadings and answers against them with the exact source passage, produces finished work product including opinions, comparative tables, risk maps and draft pleading structures, and runs a Jurimetria Agent that takes a legal thesis, analyses up to 500 selected decisions and returns tables, trend charts, divergences between courts and outcome predictions.
The company describes that agent as the first of its kind in Brazil. What keeps it off A is the A band's own test: remove the models and the corpus, the search, the collections and the jurimetric analytics remain, and those are what the company sold for the six years before GPTuri shipped. Verified 13 September 2026.
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.
Grounding is real and documented with linked primary sources and a described retrieval method, and no measured accuracy is published, which is the B band. The grounding is the product's central claim and it is specific rather than atmospheric: every answer is built on the platform's own corpus of Brazilian decisions, each analysis cites the decision it rests on, source traceability is named as a feature, and the vendor's own formulation is that the user verifies before using.
For uploaded documents the answer carries an exact reference to the passage it came from. The retrieval method is described in outline: the assistant identifies the legal elements in a natural-language description and assembles the search itself, and the vendor contrasts this with general-purpose assistants in a published comparison table covering whether the tool works from real case law, whether the full text is reachable and whether sources are traceable.
What is absent is measurement. No accuracy figure, error rate, test set or evaluation of GPTuri is published, and the one quantified claim on the estate, up to 70 per cent time saved in analysis, measures effort rather than correctness. Two disclosures cut against the product and are recorded because they are the vendor's own and a buyer should see them: the terms state that Turivius exercises no editorial control over the database, limiting itself to organising and reproducing publicly accessible data without warranty as to quality or content; and the terms state that reports, charts and decision classifications are generated automatically, may contain errors, are not exact predictions and are to be used solely as a non-exhaustive guide. Verified 13 September 2026.
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.
A written commitment that the models work alongside a supervising lawyer with real review surfaces, short of the full control structure, which is the B band. The commitment is explicit and appears in the product's own framing rather than in a disclaimer: the assistant delivers material ready for the user to review and finalise, the vendor's formulation being that you direct, it prepares, you finalise. The review surfaces behind that are real.
Each analysis cites the decision it rests on and the user is told to verify before using; answers about uploaded documents carry the exact source passage; the full text of every cited decision is reachable inside the platform; and the vendor publishes guidance of its own on validating a legal AI whose stated aim is answers that can be checked rather than answers requiring blind trust. Mode selection is also published, the user choosing GPT, Claude or Gemini per task.
What the A band asks for is missing: no threshold at which the system stops or escalates, no confidence signal, no statement of what happens after the system is wrong, and nothing describing what the Jurimetria Agent does when the evidence is thin across the up-to-500 decisions it processes. R124(2) was applied to the one candidate constraint and it does not qualify. The terms state that automatically generated reports may contain errors, are not exact predictions and must be used only as a non-exhaustive guide, which reads as the tier-and-boundary shape until the dates are checked: those terms are dated 21 September 2020 and the constraint attaches to the jurimetrics reporting that existed then, not to the generative assistant launched in 2025.
What GPTuri itself publishes is a general assurance of human review, which R124(2) holds is not enough. Verified 13 September 2026.
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.
Real deployment evidence with substance, short of joining the names to the figures, which is the B band in its own words. The naming is the strongest in this lane and it is specific. A client wall carries Deloitte, Nestlé, Gerdau, Suzano, BASF, Unimed, BTG Pactual and the law firms Tozzini Freire and Machado Meyer. Ten named case studies are published, each with its own page, covering Nestlé, Cosan and Suzano on the corporate side and Velloza, Bichara, KLA, FNCA, Bergamini, Benvenutti and GDM Consultoria Jurídica on the firm side.
Testimonials are attributed to a named individual at a named firm rather than to a role, which under R122(2) is what makes a testimonial evidence: Julia Moreira at Tortoro, Madureira & Ragazzi, Sávio Andrade at Machado Meyer, and Walter Cesar Vasconcelos at Laerte Fonseca & Advogados Associados. Third-party validation of a kind is published too, the company having won Nestlé's global innovation prize in 2022 for a project integrating jurimetrics and AI, which is notable because Nestlé is also a named customer.
What holds it off A is that the figures and the names never meet. The only outcome figure published is up to 70 per cent time saved in analysis, which is unattributed, undated and carries no method, and the reach counters on the site render as zeroes to this index because they animate client-side. The ten case study pages were not opened; under R25 they corroborate a grade that already stands on the client wall and the attributed testimonials, and they are precisely what would move this row to A if any of them joins a named client to a measured result with a stated basis. Verified 13 September 2026.
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.
Confidentiality is asserted in general terms while the specific commitments a legal buyer needs are either absent or run the other way, which is the C band. The general assertion exists: the privacy policy states that all data about the user is treated as confidential, that only authorised personnel with a need may reach it, that access to the servers is through previously authorised channels with access records kept internally, that stored information is encrypted and passwords are hashed.
That is a security posture rather than a confidentiality posture for client material, and the distinction matters on a product that invites lawyers to upload contracts, pleadings and case files. On the limbs the A band names, the record is weak and in one place adverse. Training use is permitted rather than excluded, the policy listing among its authorised purposes the use of user data in artificial intelligence models to personalise the platform, analyse use and improve services, and separately reserving the right to anonymise data submitted to the platform and use and share it in aggregate for other reasonable commercial purposes.
Segregation is not documented: the policy states that where the platform is contracted by a firm or company the employer may access the user's profile data and colleagues may see name, photograph and email, and nothing describes walls between matters, clients or users, though Projects organise material by case or client. Retention is stated only as access logs for six months and otherwise for as long as necessary. On third-party model providers the position is unstated: GPT, Claude and Gemini are named as selectable models but nothing says what any of them may retain of a prompt or an uploaded document. Privilege and professional secrecy are not addressed anywhere. Verified 13 September 2026.
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.
A real published position on advice versus tooling, short of the supervision and competence dimension, which is the B band and the common shape it names. The position is published in the terms and is more substantive than a boilerplate disclaimer. Reports, charts, decision classifications and statistical outputs are stated to be generated automatically, to be capable of containing errors, not to be exact predictions, and to be used solely as a non-exhaustive guide to jurisdictional activity.
The product framing repeats it: material is delivered for the user to review and finalise. The terms also impose a purpose limitation on the user that is unusual in this corpus and is professional-responsibility material in substance: the database and platform outputs may be used only for conducting and analysing legal proceedings, and expressly not for marketing, campaign targeting, profiling individuals, candidate selection, transferring data to third parties or business intelligence, nor for anything incompatible with the purpose for which the underlying public data was released or that would invade the privacy of the people named in it or discriminate against them.
Editorially the vendor goes further than most, publishing analyses of Brazilian courts sanctioning lawyers for fabricated citations and a guide to validating a legal AI, and connecting its own product to the verification duty those cases create. What is absent is the A band's core: no statement of who may use the product, no supervision or competence dimension, no jurisdiction limit stated though the corpus is wholly Brazilian, and no rule of professional conduct engaged. Verified 13 September 2026.
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 published governance framework with real substance, short of testing results or a named owner, which is the B band, and R36 is the ruling that puts it there: ISO 42001 is an AI management system standard and its presence alone earns B, as it did for Ontra. The About page carries an ISO 42001 mark alongside ISO 27001 and ISO 27701 and a DPO as a Service arrangement, under a heading stating commitment to the highest standards of security, privacy and artificial intelligence governance.
The limit is real and is named rather than glossed, because it is what a buyer would want to test: the mark is presented as a badge with no scope statement, no certificate number, no certifying body, no issue or expiry date and no report available, so what is established from the vendor's own surfaces is the claim to the standard rather than a verifiable audited framework. Nothing else on the estate supplies the substance a certification would imply.
No responsible AI or ethics page exists, no AI policy or acceptable use position, no accountable owner for model behaviour is named, and the DPO as a Service arrangement is a data protection function rather than an AI governance one. No testing regime is described and no evaluation result is disclosed. Bias is not addressed anywhere, which is worth naming on this product specifically: the Jurimetria Agent produces outcome predictions and success rates by court and by rapporteur, and nothing published would let a buyer test whether those predictions are even across courts, regions or parties. Verified 13 September 2026.
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 covering most of the ground, missing a stated incident practice, which is the B band and one of the two gaps it names. Retention is addressed with one specific figure and one general rule: site access records, meaning IP with date and time, are kept for six months, and everything else is kept only while necessary or relevant to the stated purposes or where law or a legitimate interest requires.
Deletion is addressed as a right, with data removed from the servers on request or when no longer necessary, subject to legal retention. Access control is described: only authorised personnel with a need may reach the data, server access runs through previously authorised channels, and records of those accesses are kept internally, with encryption at rest, hashed passwords and TLS in transit. Subprocessors are named rather than described by category, which is more than most records manage: Amazon AWS for hosting, Mixpanel, Hotjar, Google Analytics and PostHog for product analytics, and Mercado Pago, Stone, Pagar.me and PagSeguro for payments.
What is missing is incident practice. No breach notification commitment to customers was located, no notification window, no incident response description and nothing stating what a subscriber would be told or when. Two provisions are recorded because a buyer should weigh them here as well as on the confidentiality row: the policy permits use of user data in AI models to personalise and improve the service, and reserves anonymised aggregate data for other reasonable commercial purposes. Verified 13 September 2026.
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.
A real published position on liability, short of the full picture, which is the B band. What lifts this above a standard disclaimer is a warranty the vendor gives about the thing a Brazilian legal buyer would actually worry about. The terms record that to make the platform available Turivius had to collect, store and process a large body of publicly available data including judgments and other decisions, related data and legal doctrine, which may contain personal data, and that Turivius assumes full and exclusive responsibility for the legality of that database and its use, and for questions of title, originality and any breach of third-party secrecy, intellectual property or image rights, expressly relieving the user of responsibility in those situations.
That is a scoped vendor warranty with a hold-harmless attached, on the exposure the product creates by supplying scraped court data. Against it the exclusions are sweeping. Turivius, its controller, affiliates, partners and employees are stated not to be liable in any circumstance for direct or indirect damage from use of the jurimetric outputs, including outcomes that diverge from the platform's predictions, nor for direct or indirect damage, material or moral, including lost profit or revenue, arising from use of or inability to use the platform.
What the A band asks for and is not published: no liability cap of any kind is stated, no service level, no insurance, and no warranty at all attaches to the AI output as distinct from the database. Verified 13 September 2026.
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.
No integration into the systems legal work already lives in was located, which is the D band. The page inventory was taken from the navigation and footer under R20 and there is no integrations page, no developer or API documentation, no connector catalogue and no authentication model anywhere on the estate. Not one practice management system, document management system, billing platform or CRM used by Brazilian firms is named as supported, so a buyer cannot establish whether the platform connects to anything it already runs.
The product is designed as a destination rather than a layer, and the features that might look like integration on a quick read are internal and are recorded here so they are not mistaken for it: Projects hold files and saved decisions inside the platform, documents are uploaded to the platform rather than synchronised from elsewhere, and the web search complement pulls legislation and doctrine into the platform's own context rather than pushing anything out.
The one outbound provision located is contractual rather than technical: the terms permit the user to present platform-generated reports to third parties, individually and without altering their format. Nothing describes exporting structured data, and the terms separately prohibit running any program to scrape or index the platform, which forecloses the route a firm might otherwise take. Verified 13 September 2026.
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.
Cloud delivery with an express residency commitment and no tenancy model described, which under R38 is B because tenancy and region are co-equal limbs and publishing either clears C. Residency is stated on the home page as one of three named security commitments: data storage exclusively in Brazil, with the information remaining in national territory on trusted infrastructure. Alongside it the vendor publishes full LGPD compliance and encrypted traffic over HTTPS, SSL and TLS.
The privacy policy identifies the hosting arrangement, stating that all data provided is stored on servers external both to the user's employer and to Turivius, for example on Amazon AWS, and directing the reader to that provider's own policy. What is not described is tenancy. Nothing states whether a firm's uploaded documents and saved collections sit in a shared or isolated environment, no separation model is published, and there is no on-premises or private deployment option offered or refused.
One tension is named rather than smoothed, because a buyer evaluating the residency claim needs it: the storage commitment is written about storage, and the AI environment sends the user's prompts and uploaded documents to GPT, Claude or Gemini, none of which is a Brazilian service. Nothing on the estate reconciles the two, states where inference runs, or says whether anything leaves the country in the course of generating an answer. Verified 13 September 2026.
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.
Badges appear on the site with no scope, no date and no report available, which is the C band word for word. The About page carries marks for ISO 27001, ISO 27701 and ISO 42001, together with a DPO as a Service arrangement, under a heading stating commitment to the highest standards of security, privacy and artificial intelligence governance. That set is a serious one, covering information security, privacy information management and AI management, and the AI standard in particular is held by very few vendors in this corpus.
What cannot be established from the vendor's own surfaces is anything behind the marks. No certificate number, no certification body, no issue or expiry date, no statement of the certified scope, no surveillance audit record and no report obtainable by any route, gated or otherwise. There is no trust centre, no security page and no compliance page anywhere in the navigation or the footer. The heading's own wording is commitment to standards rather than certification against them, which leaves a buyer unable to tell whether the marks record completed audits or an alignment programme.
What is published in place of evidence is a short set of assertions on the home page, being LGPD compliance, storage exclusively in Brazil and encrypted traffic, and a longer description of controls in the privacy policy covering authorised access, internal access logging, encryption at rest and hashed passwords. Those are the vendor's own statements, not independent attestation. Verified 13 September 2026.
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, providers named without change notification, which is the B band exactly as written. The naming is unusually direct and it is a purchasable feature rather than a compliance footnote: GPTuri offers GPT, Claude and Gemini in one interface and the buyer selects which runs each task, with the vendor publishing a short characterisation of each, versatility and speed for everyday analysis and synthesis, deep reasoning for complex analysis, long contracts and strategic argument, and extended context across multiple documents.
All three are stated to read the same corpus of more than 130 million decisions. So a buyer learns which model families process its work and can choose between them, which is more control than any other record in this pull offers. Hosting is identified separately in the privacy policy, which names Amazon AWS as the external server platform. What the A band requires and is not published: no model version is named for any of the three, so a buyer cannot establish which release is running or when it changes; the provider entities behind the product names are never written out; nothing states where inference runs, which matters against the home page's commitment to store data exclusively in Brazil; nothing states what any model provider may retain of a prompt or an uploaded document; and no commitment to notify customers of a change to any of it was located. Verified 13 September 2026.
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.
Pricing is gated behind a demo request while the tier names and the feature split are published, so the shape is visible and the number is not, which is the C band word for word. The shape is genuinely visible. Two environments are named and differentiated on the vendor's own surfaces, Turivius Clássica covering jurisprudence, jurimetrics and knowledge management and GPTuri+ covering advanced legal AI for research, documents and jurimetrics, sold as independent accesses with separate logins, and the feature content of each is described at length across dedicated pages.
The unit and the commercial mechanics are published in the terms, which is more than the C band assumes: the subscription is monthly, renewing automatically every thirty days, the licence is personal and non-transferable with password sharing prohibited and grounds for suspension, cancellation may be made at any time and takes effect at the end of the paid month with no refund for the unused remainder, and the vendor reserves the right to change the price at any time while guaranteeing the rate in force on the first day of the current billing month.
What is absent is any figure. The terms direct the reader to the site or to email for prices; the site's every purchase path, including the button labelled create an account, resolves to a single page headed schedule your demonstration, which is a booking form carrying client logos and no rate card. No entry price, no band, no per-seat rate and no minimum is published at any level, and enterprise arrangements are contracted separately under the terms' own carve-out. Verified 13 September 2026.
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.
Segment coverage is described with substance and the boundaries are left open, which is the B band. The two buyer segments are addressed separately and in their own terms rather than as a single audience. For law firms the published propositions are faster research, more relevant results, centralised organisation of theses and collections, and quicker delivery to clients. For in-house legal departments they are more reliable data, clearer reporting that translates legal complexity for the business, less operational work in favour of strategy, and faster internal responses to other functions.
The named customer base evidences both halves, running from Deloitte, Tozzini Freire and Machado Meyer on the firm side to Nestlé, Gerdau, Suzano, BASF, Unimed and BTG Pactual in-house. Jurisdictional coverage is stated with a figure, more than 100 courts across Brazil and more than 130 million decisions, updated daily, and the company describes jurimetrics as covering the principal areas of law. What is left open is everything the A band asks for at the edges.
No practice area is named as supported or unsupported, and the claim to cover the principal areas of law is exactly the unevidenced breadth claim the band below is written for, saved from it by the court and volume figures that are specific. Nothing states which courts or which years the corpus reaches, whether coverage differs between the two environments, or whether the Jurimetria Agent works across the whole corpus. Jurisdiction is Brazil throughout and is never stated as a limit. Verified 13 September 2026.
5 public documents
The public pages on file for Turivius, with the recorded signals each one supports and the date it was last read. Open any of them and check the reading against the record.
-
turivius.com/politica-de-privacidade4 signals
Client Data in Training, Prompt and Output Retention, Ethical Walls and Matter Segregation and 1 more
Read Sep 13, 2026
-
turivius.com/termos-de-uso3 signals
Primary Law Corpus Provenance, Bar Guidance Alignment, Court Disclosure Support
Read Sep 13, 2026
-
turivius.com/gpturi2 signals
Refusal and Uncertainty Behaviour, Outside Counsel Guideline Readiness
Read Sep 13, 2026
-
turivius.com/checkout1 signal
Billing and Fee Posture
Read Sep 13, 2026
-
turivius.com/jurimetria1 signal
Good Law Verification
Read Sep 13, 2026
No published figure
- Turivius publishes what you would be buying and none of what it costs.
- There are two environments and they are sold and logged into separately. Turivius Classica is the established product: case law search across a database the company puts at more than 130 million decisions from over 100 Brazilian courts, plus jurimetrics and tools for organising decisions into collections by thesis, client or case. GPTuri+ is the AI environment, with the generative assistant, document upload and questioning, the choice of GPT, Claude or Gemini per task, and the Jurimetria Agent. The feature content of each is described at length, so you can tell what the difference is.
- What you cannot find is a price. There is no rate card, no entry price, no per-seat figure and no published band, for either environment. Every route that looks like it leads to a purchase, including the button that says create an account, takes you to the same page asking you to book a demonstration. The terms tell you to check the site or email them, and the site tells you to book a demo.
- The commercial mechanics are published, and they are worth reading before you commit. The subscription is monthly and renews automatically every thirty days. The licence is personal and non-transferable, so sharing a login is a breach and grounds for suspension. You can cancel at any time, but you keep access only to the end of the month you have already paid for and there is no refund for the unused part. And the company reserves the right to change the price at any time, with your existing rate protected only until the end of the current billing month.
Published tier structure and charging mechanics with no figure at any level. Two separately sold environments with independent logins: Turivius Classica, covering jurisprudence search, jurimetrics and legal knowledge management, and GPTuri+, covering the generative assistant, document analysis, multi-model selection across GPT, Claude and Gemini, web search complement, voice input and the Jurimetria Agent. The feature content of each is described across dedicated product pages, so the shape of what is bought is visible.
The unit and mechanics are published in the Terms of Use: the subscription is monthly, renewing automatically every thirty days at the price then indicated; the licence is non-exclusive, limited, revocable, onerous, personal and non-transferable, with password sharing prohibited and grounds for suspension without indemnity; cancellation may be made at any time and is treated as notice not to renew, with access continuing to the end of the paid month and no restitution for the unused remainder; and the vendor reserves the right to change the price at any time, guaranteeing the rate in force on the first day of the current billing month.
Suspension is reserved for suspected fraud or non-payment. A demonstration may be granted at the vendor's discretion, free for an agreed period and withdrawable at any time without notice or indemnity. Enterprise arrangements sit outside all of this under an express carve-out for terms agreed in a separate contract. No entry price, band, per-seat rate, minimum commitment, term length or implementation charge is published on any surface, and every purchase path on the site resolves to a demonstration booking form.
Confidentiality and data terms: No data processing addendum, business associate agreement or separately signable data instrument is published. The data position lives in a privacy policy last updated 27 July 2023 and in Terms of Use dated 21 September 2020, both of which predate the generative assistant launched in 2025, and neither contains an AI clause drafted for it. What the policy does contain on the point is a permission rather than a protection: user data may be used in artificial intelligence models to personalise the platform, analyse its use and improve the services, and data submitted to the platform may be anonymised and used or shared in aggregate for other reasonable commercial purposes. Processors are named, which is more than most records manage, being Amazon AWS for hosting, Mixpanel, Hotjar, Google Analytics and PostHog for analytics and four Brazilian payment providers. Security is described as authorised access only through previously authorised channels with internal access records, encryption at rest, hashed passwords and TLS in transit, alongside home page commitments to full LGPD compliance, storage exclusively in Brazil and encrypted traffic. Certification marks for ISO 27001, ISO 27701 and ISO 42001 appear on the About page with no scope, date, certificate number, certifying body or obtainable report, and there is no trust centre. No breach notification commitment to customers was located.
Note: Tier structure, feature split and commercial mechanics read from the vendor's own pages and Terms of Use on 13 September 2026. entryPriceUsd is null because no figure is published anywhere on the estate at any level, and entryPriceDisplay is empty per R10 and R19, which reserve the display field for a figure or nothing. Currency is not set because no charging currency is stated, though the contracting entity is Brazilian and the forum is Sao Paulo. The row is written under R17 rather than R10: the published tier names, feature split and charging mechanics lift Commercial Transparency above D to C, and the mechanical test is that where pricing evidence lifts the axis off the floor a row is owed even with no figure. It is the Bloomberg Law shape, structure and unit published and the number withheld. Recorded so the C is read correctly rather than as a retrieval limit: this is not a page that would not load. The checkout path resolves cleanly and was read in full; it is a demonstration booking form carrying client logos, and every purchase route on the site, including the button labelled create an account, leads to it. The terms themselves direct the reader to the site or to email for prices, so the two published sources point at each other. Enterprise arrangements are contracted separately under the terms' own carve-out for a bespoke contract, so nothing is published for the top of the range either.
Legal Signals
What each signal meansA 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.
Client Data in Training
Can material a lawyer puts into this product be used to train a model?
Public material states that customer content trains, refines or personalises models, with no matching term located in the published agreement. Any de identification, anonymisation or aggregation qualifier is recorded in the summary.
Public material states that user data is used in AI models to personalise and improve the service, with no matching term in the published agreement, which is this value exactly as R43(c) defined it. The privacy policy lists among its authorised purposes the use of user data in artificial intelligence models, stated to be only for the purposes set out in that policy and in particular to personalise the platform, analyse its use and improve the services.
That is a permission rather than a prohibition and it is written in the vendor's own voice. A second clause widens it: the policy reserves the right to anonymise data submitted to the platform and to share and use it in aggregate for other reasonable commercial purposes, never permitting identification. On a product that invites lawyers to upload contracts, pleadings and case files, data submitted to the platform is client material.
R43(1) was run and is discharged. The Terms of Use were located and read in full and contain no AI clause of any kind, so no contractual term matches the policy permission and the value is policy rather than contractual. Two dates belong on the record because they explain the shape of it: the Terms are dated 21 September 2020 and the privacy policy 27 July 2023, while GPTuri, the generative assistant this signal is really about, launched in 2025.
The instruments governing the AI predate the AI, and the only sentence in either that reaches model training is a permission written before the product existed.
Prompt and Output Retention
How long does the product keep what a lawyer typed, and can that be set to zero?
Retention is acknowledged in public materials with no stated period.
Retention is addressed and no period is stated for the material this signal covers, which is this value. One specific figure is published and it is not the relevant one: site access records, being IP address with date and time, are held for six months under the Brazilian Civil Rights Framework for the Internet. Everything else is governed by a general rule, that data is kept only for as long as it is necessary or relevant to the purposes described, or for periods fixed by law, or while a legitimate interest of the vendor requires it.
Nothing narrows that for prompts, uploaded documents or generated outputs, and on this product those are the material that matters: the assistant takes contracts, pleadings, notifications and expert reports uploaded by the user, and Projects are designed to retain files, saved decisions and analyses so that they can be carried into later conversations, which is a retention feature sold as a benefit and never given a period.
No deletion-on-termination commitment was located and no export or return obligation. Deletion is addressed only as a data subject right exercisable on request, with the vendor reserving grounds to keep material where a legal duty or its own legitimate interest applies. Nothing states what any of the three model providers may retain of a prompt or an uploaded document once it has been sent to them.
Ethical Walls and Matter Segregation
Does retrieval respect the firm’s ethical walls, or can the model read across them?
Segregation is asserted in public materials with no published detail on how it is enforced.
Access control is claimed and no permission model is described, which is this value. What is claimed is vendor-side and general: all data is treated as confidential, only people inside Turivius with the appropriate authorisation and a need may reach it, access to the servers runs through previously authorised channels, and records of those accesses are kept internally. That is an internal access statement rather than a segregation model for the customer's own material.
On the customer side the only published statement runs the other way and is recorded because a buyer should see it: where the platform is contracted by a company or a law firm for use by its employees or lawyers, the policy states that the employer may have access to the user's profile data and that other staff of the same organisation may see the user's name, photograph and email. Profile data is defined broadly enough to include search history and platform usage.
Nothing describes walls between matters, between clients, or between users within a subscribing organisation. Projects organise files, decisions and analyses by case or client, which is the natural place a matter boundary would live, and nothing published says whether a Project is private to its creator, visible to the account administrator, or shared across the organisation, nor how access to one is granted or withdrawn. No role model, no permission levels and no conflicts handling were located.
Third Party Request and Subpoena Notice
If someone subpoenas the vendor for a firm’s data, does the firm hear about it first?
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 and customer notice is absent, which is this value. The privacy policy reserves the right to supply data and information where judicially requisitioned, expressly limited to the extent the disclosure is legally obligatory, necessary for the company to comply with national law, or expressly authorised by the user. The application access records the vendor is statutorily required to collect are governed by a tighter rule that is worth recording because it is narrower than most in this corpus: those records will be provided to third parties only with the user's express authorisation or under judicial demand.
Both provisions are about the conditions of disclosure. Neither says anything about telling the customer. Nothing states that the subscriber or the user would be notified that a demand had been received, given an opportunity to object or to seek relief before production, or informed after the event, and no transparency report or law enforcement guidelines page exists on the estate. The Terms of Use, read in full, contain no provision on legal process at all.
One adjacent clause is recorded and not credited because it concerns a different route out: the policy states that user data may be treated as a company asset and transferred if the business is sold, acquired or merged, with the user's agreement taken as given by the policy itself.
Primary Law Corpus Provenance
Where does the law in this product come from, and does the vendor have the right to use it?
Sources are identified without stating the licence or rights basis.
The sources are named and no licence is identified, which is this value, though the clause carrying it is stronger than the value implies and the note records why. The Terms describe the corpus directly: to make the platform available Turivius collected, stored and processed a large body of publicly available data, being data on legal proceedings such as first instance judgments, appellate decisions and other rulings, related data, legal doctrine, and other data gathered by Turivius or by its partners, which may include personal data.
The vendor then assumes full and exclusive responsibility for the legality of that database and for its use in the platform, and for questions of title and originality and any breach of third-party secrecy, intellectual property or image rights, expressly relieving the user of responsibility. A separate clause states that the vendor exercises no editorial control over the database, limiting itself to organising, structuring and reproducing publicly accessible data without warranty as to quality or content.
So the provenance is stated, the classes of source are identified, and the legal risk is allocated to the vendor. What is absent is a licence. The corpus rests on public availability rather than on any identified agreement with a court or publisher, no supplier is named, the partners referred to are not identified, and nothing states on what basis third-party doctrine sits in the database. Nothing addresses whether the public-availability basis extends to processing the corpus with third-party models.
Good Law Verification
Does the product tell you when the authority it just cited has been overruled?
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 note separates that from the analytical work the platform genuinely does. What is published is quantitative rather than status-based. The jurimetrics layer reports rates at which claims succeed, the most cited grounds, divergences between courts and the behaviour of individual rapporteurs, and the Jurimetria Agent identifies sub-theses across up to 500 selected decisions and returns trend charts and outcome predictions.
Divergence between courts is a description of how a question is being decided in different places, not a statement that a given decision has been overruled, superseded or is no longer good law. No flag, indicator, treatment status or currency signal of any kind was located, and no editorial or algorithmic process for establishing treatment is described. The vendor's own terms cut against any implied assurance and are recorded here because they are the clearest statement on the point: Turivius states that it exercises no editorial control over the database, limiting itself to organising, structuring and reproducing publicly accessible data, with no warranty as to its quality or content.
On a platform whose assistant selects precedent for lawyers to rely on, the currency of that precedent is unaddressed. The surfaces read on the date shown were the home page, the GPTuri and Jurimetria pages, the terms, the privacy policy, the cases index and the vendor's own blog on validating legal AI.
Refusal and Uncertainty Behaviour
What does the product do when the answer is not in the corpus?
No located public material addresses what the product does when it cannot ground an answer.
No located public material describes what the system does when it cannot produce a reliable answer. The vendor addresses the risk at the level of architecture and of user duty rather than system behaviour, and both are recorded because they are substantive and neither is this signal. Architecturally, every answer is built on the platform's own corpus with the cited decision reachable in full, and the vendor's published position is that the aim is not answers requiring blind trust but answers that can be checked.
As a matter of user duty, the terms state that automatically generated reports and classifications may contain errors, are not exact predictions and must be used only as a non-exhaustive guide. What neither does is say what happens inside the product at the moment of uncertainty. Nothing states that the assistant declines a question it cannot ground, reports that it found no supporting decision, attaches a confidence signal, flags a weakly supported passage, or behaves differently when the Jurimetria Agent finds few or conflicting decisions across the sample it analyses.
The outcome predictions the agent produces are presented without any published statement of when the underlying evidence is too thin to predict from. The surfaces read on the date shown were the home page, the GPTuri, Jurimetria and About pages, the terms, the privacy policy and the vendor's own published guidance on validating a legal AI.
Fabricated Citation Record
Does a public court record exist addressing fabricated or hallucinated legal citations in output from this product?
No court order, opinion or disciplinary record addressing fabricated or hallucinated legal citations produced by this product has been located as of the date shown. This is a statement about the public record on that one subject, not a finding about the product, and this signal is not a litigation history.
Searched on 13 September 2026 against the company name and the product name in Portuguese and English, across Brazilian legal press and court reporting on AI-generated fabricated citations. None located. No decision, sanction or disciplinary referral names Turivius or GPTuri. The context is recorded because Brazil now has an active body of such cases and a reader should see that the absence was tested against it rather than against an empty field.
Reported instances include the Sixth Panel of the Superior Labour Court imposing a bad-faith litigation fine on a telecommunications company and its lawyer in March 2026 after non-existent case law was cited in appellate submissions and detected by the reporting judge's own chambers; a Santa Catarina State Court appeal in which false precedents with indications of AI use drew a bad-faith penalty; federal regional court decisions imposing fines and, in one case, referring the lawyer to the Bar; a Minas Gerais labour court fine for an invented binding summary; and a labour court treating an unverified AI-drafted initial petition as a procedurally non-existent act.
General-purpose assistants rather than legal platforms are what those accounts describe. One point of interest rather than of grading: the vendor publishes its own analysis of several of these decisions on its blog, prepared using GPTuri. Under R119 this signal records fabricated legal citations in filings and nothing else.
Bar Guidance Alignment
Has the vendor engaged in public with the ethics opinions its buyers are bound by?
Public materials refer to professional responsibility in general terms without naming guidance.
Professional responsibility is engaged in general terms with no authority named, which is this value, and the engagement is more active than most records at this level. Three things sit behind it. The terms state that automatically generated reports, charts and decision classifications may contain errors, are not exact predictions and are to be used solely as a non-exhaustive guide to jurisdictional activity, which is a duty pointed at the right question.
The terms also impose a purpose limitation on the user that is professional-responsibility material in substance, confining use of the database to the conduct and analysis of legal proceedings and expressly forbidding profiling of individuals, candidate selection, marketing and campaign targeting, business intelligence, transfer to third parties, and any use incompatible with the purpose for which the underlying public data was released or that would invade privacy or discriminate.
And editorially the vendor engages the duty directly, publishing analyses of Brazilian courts sanctioning lawyers for fabricated citations, including referrals to the Bar, alongside guidance of its own on how to validate a legal AI, and connecting its product to the verification duty those decisions create. What is absent is any named authority. No provision of the Brazilian Bar's statute or code of ethics is cited, no Bar opinion or resolution is referenced, no court guidance is mapped to the product, and the Bar appears on the estate only as a body to which other lawyers were reported.
Billing and Fee Posture
Does the vendor address what happens to the bill when the work takes an hour instead of six?
The product sits inside a lawyer to client fee relationship and no located public material addresses billing, fee or disclosure treatment, with no savings claim published either.
Nothing published addresses what happens to the bill when AI-assisted work takes an hour instead of six, which is the floor, and all the higher values are false of this record. The product sits squarely inside a lawyer-to-client fee relationship: it is sold to law firms and to in-house departments, the firm-side value proposition is explicitly about faster research and quicker delivery to clients, and the single quantified claim on the estate is up to 70 per cent time saved in analysis.
That claim is about the practitioner's time, and time is what a Brazilian firm bills. Nothing follows from it. No per-matter record of AI-assisted work is described, nothing marks an output as machine-generated for the purposes of a bill or a fee note, no guidance on fee or disclosure treatment is published, and nothing addresses whether a subscription cost or a per-analysis cost may be passed to a client as a disbursement.
Recorded and expressly not credited under R21 and R24, because software cost is a different object from AI-assisted work: the terms publish the charging mechanics, being a monthly subscription renewing every thirty days with no refund for an unused remainder, which would let a firm attribute the platform cost to a period but says nothing about the client's bill. All the higher values being false of this record, this is a gap rather than a grading error and the summary carries it.
Outside Counsel Guideline Readiness
Can a firm get this vendor through a client’s AI clause without a bespoke negotiation?
A current subprocessor or model provider list is published.
A processor list and a model statement both exist and no forwardable client-facing pack sits around them, which is this value under R29's IPRally condition. On the model limb the vendor is more forthcoming than most: the product page states that GPT, Claude and Gemini are available in one interface, that the buyer selects which runs each task, and that all three read the same corpus, so a firm can tell its client which model families process its work and can choose between them.
R29's first test is passed because this is model naming rather than infrastructure naming. On the processor limb the privacy policy names Amazon AWS as the hosting platform, Mixpanel, Hotjar, Google Analytics and PostHog for product analytics, and Mercado Pago, Stone, Pagar.me and PagSeguro for payments. What is missing is the third limb and some of the substance behind the first two. There is no data processing addendum, no consent or notification pack drafted to be forwarded, and no client-facing disclosure artifact of any kind.
The model naming lives on a marketing page rather than in a disclosure instrument, gives no provider entity, no version and no statement of what any model provider may retain, and is not reconciled anywhere with the home page's commitment to store data exclusively in Brazil. R29 is explicit that where the list sits outside a DPA and no other forwardable artifact exists, the value drops rather than the top two values collapsing into each other.
Court Disclosure Support
If a judge’s standing order requires an AI disclosure, can the product produce one?
No located public material addresses court disclosure or verification certification.
No located public material addresses disclosure of AI involvement in legal work, which is the floor. One clause comes closer than most and is recorded and not credited, because on a careful reading it points the other way. The terms expressly permit the user to present platform-generated reports to third parties, individually and without altering their format or image. That is a permission to circulate the output, and a Brazilian court is a third party, so a jurimetric report can properly be put in front of one.
What the clause does not do is make the output self-identifying: nothing states that a report carries a mark, a header or any indication that it was machine-generated, nothing records which model produced it or when, and the format-preservation condition protects the vendor's presentation rather than the reader's ability to tell what they are looking at. Beyond that clause there is nothing. No audit trail of AI use, no per-query or per-matter record a firm could produce, no certification template, and no guidance on whether or how AI assistance should be disclosed to a court or a client.
The gap has weight in this jurisdiction: Brazilian courts have sanctioned lawyers for filings containing fabricated citations and have treated an unverified AI-drafted petition as a procedurally non-existent act, so the ability to evidence how an output was produced is becoming a practical need rather than a theoretical one.