Trellis

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.

Machine learning is what makes the underlying record set usable, on a platform whose base offering is search over public court data. The vendor describes Trellis AI as combining its data foundation with advanced language models to produce case assessments that examine facts, claims and defences and return potential outcomes, recommended actions and risk factors, alongside argument drafting and generation. Underneath that sits a docket search engine over 3,000-plus courts, which is close to what the company sold at founding when it indexed California Superior Court records. Remove the models and a searchable trial court database remains, which is the B band rather than the A.

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.

Precision is asserted in marketing and disclaimed in the agreement. The vendor describes Trellis AI as leveraging advanced language models to deliver precise insights that enhance decision-making and case preparation, and positions the platform on unmatched coverage. The terms of service then disclaim liability for the omission or inaccuracy of any court-provided data and for any content, errors in or omissions from the online services. No accuracy figure, error rate, benchmark, test set or published evaluation was located on any surface read on 31 Aug 2026. That gap matters more here than on a drafting tool, because the product's output is factual assertions about how named judges have ruled, and a wrong one is not detectable by a reader.

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.

Autonomy is claimed and no oversight mechanism is published. Trellis AI is described as examining case facts, legal claims and defences to deliver insights into potential outcomes, recommended actions and risk factors, with each assessment delivering actionable intelligence and recommended next steps to inform decisions on case management, settlement strategy and trial preparation, alongside argument drafting and generation. Nothing published states what the system produces unattended, where a lawyer must review before relying on an assessment, or what happens when an output is wrong. Searched the home page, the vendor's product announcement, the API documentation, the knowledge base index, the terms of service and the privacy policy on 31 Aug 2026. A vendor recruitment listing describes contract attorneys labelling verdict types and monetary awards through custom software to ensure data accuracy, which suggests human review in the data pipeline, but a job advert is not product material and does not carry a grade.

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.

Named users at named institutions, with scale figures that carry no customer. Attributed quotes come from Brian Cassidy, Student Services Librarian at Cleveland State University Law Library, on strategic trial court comparison analytics; Dean Walters, Assistant Director for Content at Harvard Law School, on the platform giving empirical sourcing to what was once anecdote; and Miguel Aristizabal, Partner at Clayton Trial Lawyers. All three are qualitative. The vendor separately states it serves tens of thousands of law firms and litigators daily across 3,000-plus courts in over 2,500 counties spanning 45 states. No figure for what changed at any named customer was located, and no case study with a stated method was found on the surfaces read on 31 Aug 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.

The vendor makes an explicit confidentiality claim, and the rest of its own material sits against it. The claim, published on the Trellis blog, is that data confidentiality is a top priority, that the platform relies solely on publicly available legal filings, that this eliminates the need for users to input sensitive client or internal data, and that Trellis does not handle personally identifiable information or any confidential case details. That is a real mitigating position and it is why this is not lower. Three things on the other side, all from vendor material. The privacy policy describes collecting personal information and limits disclosure of usage data to unaffiliated third parties except as necessary to service the account, enforce the terms, meet obligations to content and technology providers, or as required by law, which is a general assurance rather than a commitment on what the product holds. The knowledge base states that judge pages carry career history and political affiliation, so the platform does handle identifying information about named individuals, albeit public officials rather than customers. And Trellis Envelopes is described as an in-app messaging service for sharing rulings, dockets and documents between users, which is a channel for exactly the internal material the claim says users never need to input. The graded position is the documented one: no statement on whether user queries or shared material train any model, no retention period, no segregation or access model, and no privilege or work product treatment, on a product where what a litigator searches reveals strategy. Checked 31 Aug 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.

Nothing published on the advice line. Searched the home page, the vendor product announcement, the knowledge base index, the terms of service and the privacy policy on 31 Aug 2026. No statement that Trellis output does not constitute legal advice, no professional responsibility or ethics page, no bar or ethics guidance named including ABA Formal Opinion 512, and nothing addressing a lawyer's competence or supervision duties when relying on a case assessment that returns recommended actions and settlement guidance. The terms do carry one significant published use restriction, but it is regulatory rather than professional: no user may use the data to determine a consumer's eligibility for credit, insurance or a government licence or benefit, with consumer defined by reference to the Fair Credit Reporting Act at 15 USC 1681. That addresses the vendor's exposure as a data provider, not the lawyer's obligations.

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.

No governance position for model behaviour was located. Searched the home page, the vendor's Trellis AI product announcement, the API documentation, the knowledge base index, the terms of service and the privacy policy on 31 Aug 2026. Nothing names an accountable owner, describes pre-release testing, publishes a responsible AI framework, or discloses anything about uneven output. The absence is conspicuous on this product in particular: the platform builds analytical profiles of named individual judges, and the knowledge base states that judge pages carry career history and political affiliation alongside ruling tendencies. A product that scores identifiable public officials and surfaces their political affiliation is the clearest case on this index for published bias testing, and none exists.

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.

A general privacy policy covers the website and account relationship without addressing what happens to what a user does inside the product. The policy states that Trellis does not sell customer or user data and limits disclosure of usage data to unaffiliated third parties, and that a user may cancel registration at any time through settings. Searched the home page, features page, product announcement, API documentation, knowledge base index, blog, terms of service and privacy policy on 31 Aug 2026 and located no security page of any kind: no encryption standard, no access control model, no retention period, no named subprocessor, no incident or breach notification practice, and no deletion commitment beyond account cancellation. The vendor does argue the exposure is low, stating in its own blog that the platform relies solely on publicly available filings and does not handle personally identifiable information or confidential case details, which is a real mitigating position but sits in tension with a privacy policy that describes collecting personal information and with an in-app feature for sharing documents between users.

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.

Liability is addressed only through a limitation clause, and that clause disclaims precisely the exposure the product creates. The terms of service state that Trellis shall not be liable for any loss, injury, claim, liability or damage of any kind resulting in any way from any content, errors in or omissions from the online services, the unavailability or interruption of the services, the customer's use of them, the loss or corruption of any data, or the omission or inaccuracy of any court-provided data, all to the fullest extent permissible by applicable law. The services may also be enhanced, added to, withdrawn or otherwise changed without notice. No indemnity running to the customer, no warranty on output and no insurance position was located, and the cap was not among the text retrievable on 31 Aug 2026. A separate subscription agreement is referenced by the terms and was not located, so the full allocation may sit there.

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.

One real integration route, documented at a high level, and nothing else. Trellis publishes an API with its own support documentation, described as providing access to state and federal court data that was previously siloed and fragmented and as available for integrating Trellis directly into a customer's workflow. That is a genuine route for a firm building its own tooling. What was not located on 31 Aug 2026 is integration into the systems litigation work otherwise lives in: no document management, no matter management, no practice management, no filing system and no named third-party connector of any kind. Nor was implementer-level documentation reached, since only the API overview was retrievable and it describes availability rather than what the endpoints return or how authentication and rate limits work.

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.

Nothing published on where the software runs or where customer data sits. Searched the home page, the product announcement, the API documentation, the knowledge base index, the terms of service and the privacy policy on 31 Aug 2026. No hosting provider is named, no tenancy model is stated, no region or residency option is offered, and nothing distinguishes processing from storage. The single geographic statement located runs the other way and is a restriction rather than an option: the terms of service state that, to comply with local privacy, data protection and other laws, the customer may not access Trellis outside the United States. That tells a buyer where they may use the product, not where their data lives.

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.

No independent security attestation was located in any vendor material. Searched the home page, the features page, the vendor's Trellis AI product announcement, the API documentation, the knowledge base index, the vendor blog, the terms of service and the privacy policy on 31 Aug 2026, and ran a targeted search for a Trellis security or trust page. No certification is named, no auditor identified, no scope or coverage period given, no penetration test referenced and no trust centre or portal exists. A third-party blog asserts that the company follows SOC 2 security standards, but no vendor-published material corroborates it and an unverified secondary source cannot support this axis. One trap worth recording for any future grader: a different and unrelated company also trading as Trellis, at trelliscare.app, publishes a detailed healthcare security page covering HIPAA, business associate agreements and zero data retention. It is not this vendor and must not be read across.

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.

Nothing published about the model supply chain a customer inherits. The vendor's own product announcement states that Trellis AI leverages its data foundation along with advanced language models, which acknowledges that language models are involved and identifies none of them. Searched the home page, that announcement, the API documentation, the knowledge base index, the terms of service and the privacy policy on 31 Aug 2026: no provider is named, no model or version is identified, no architecture is described, no processing location is given, no subprocessor list exists, and nothing commits the vendor to notifying customers when any of it changes. The privacy policy's reference to obligations to content and technology providers acknowledges third parties in the chain without naming one.

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.

Written 31 Aug 2026 as an R7 amendment. The original build left this unwritten rather than graded D, because trellis.law/plans was bot-blocked and the only figures available came from third-party directories that contradicted each other. That hold was correct: D would have asserted no pricing is published at any level, and the vendor publishes a great deal. The figures were recovered from Trellis's own knowledge base at support.trellis.law/what-are-the-different-tiers, fetched 31 Aug 2026, which sets out four subscription plans. Three carry published monthly and annual rates: Personal at $69.95 a month or $649.95 a year, Research at $129.95 a month or $1,099.95 a year, and Research and Judge Analytics at $199.95 a month or $1,999.95 a year. The unit is stated as well as the price, each tier carrying an annual content view allowance, 240 on Personal and 900 on the two Research tiers, with single state coverage on all three and Judge Analytics as the feature that separates the top published tier. The fourth plan, Law Firm and Academia, has no figure and routes to a sales contact form. Document access is a separate axis of charge, with some documents included in the subscription, some requestable at no additional cost in Los Angeles and Cook counties within a monthly limit, and others purchasable, and an enterprise deferred billing add-on exists for document purchases. That is real pricing across part of the range with the enterprise tier withheld, which is the B band. It falls short of A because the firm-level tier a serious buyer would actually purchase is the one without a number, and because nothing states what implementation or onboarding adds.

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 quantified precisely and the audience is segmented in detail, with the boundaries stated only in part. The vendor publishes coverage as more than 3,000 courts across over 2,500 counties spanning 45 states, and the API documentation and plans navigation add federal circuits, bankruptcy courts and the Supreme Court. Audience segmentation runs to role rather than just firm type, with published pages for associates, partners, knowledge management and marketing or business development inside law firms, plus in-house legal, and the customer evidence adds academic law libraries. Practice coverage is expressed concretely through motion type and legal issue pages and through published treatises and primers on how common motions are handled at trial court level. Two limits are stated: access is restricted to the United States by the terms of service, and the data may not be used for Fair Credit Reporting Act eligibility determinations. What is not stated is which of the 45 states are covered to what depth, and the state-only positioning used in much of the marketing sits awkwardly beside the federal coverage the API page claims.