Trellis AI Review: Litigation Analytics for Practicing Lawyers
Trellis sits in a niche that most lawyers didn't know existed until recently: normalized, searchable state trial court data. Federal court records have PACER (flawed as it is); state trial dockets have historically been a mess of county-by-county portals with inconsistent search, inconsistent redaction, and no analytics layer. Trellis is one of a small group of vendors trying to fix that, and it now layers generative AI on top of the underlying docket data.
This review is aimed at litigators, insurance defense lawyers, plaintiffs' firms doing venue analysis, and general counsel doing outside-counsel diligence. I'll cover what Trellis actually does, the confidentiality picture (which for a research tool is different than for a notetaker or a drafting assistant), how it lines up with the ABA's July 2024 guidance on generative AI, pricing, and where it fits — and doesn't fit — in a modern practice stack.
What Trellis Actually Does
At its core, Trellis is a state trial court research and analytics platform. According to the vendor's site, the platform aggregates state trial court records from courts across the United States and provides search, docket monitoring, judge analytics, and case tracking on top of that data. That is a materially different offering than Westlaw or Lexis, which are strong on appellate authority and secondary sources but historically thin on state trial court filings and outcomes.
The product has several layered use cases:
- Judge analytics. How a specific state trial judge tends to rule on categories of motions, timing patterns, and the types of cases on their docket.
- Opposing counsel and party research. Prior filings, case volumes, and litigation history for a firm, lawyer, or repeat-player party.
- Docket alerts. Monitoring of new filings matching saved criteria, useful for business development and for tracking active cases.
- Document retrieval. Pulling the underlying complaints, motions, and orders from state trial courts that would otherwise require a runner or a county-portal fishing expedition.
- Generative AI features. Trellis has added AI-assisted summarization and Q&A over case documents. See Trellis AI for the current feature set.
The strongest, most defensible piece of the offering is the underlying data. If you practice in a jurisdiction where the state court e-filing system is punishing to search — California superior courts are the canonical example — a tool that gives you clean, cross-county search over complaints, motions, and orders is a real productivity gain. That is true even before you touch any AI layer.
The Confidentiality Lens
Confidentiality analysis for a research tool looks different than for a notetaker or an email drafting assistant. You are not, in the normal case, uploading privileged client communications into Trellis. You are typing queries and reading public court records. But that distinction can lull lawyers into a false sense of security, so it is worth being precise.
Three data streams matter:
- Your queries and search history. These are metadata about your practice. "Who else has sued my client's supplier in Harris County over the last 24 months" is a query that, on its face, tells the vendor something about your matter. Aggregated over months, your search log is a fair proxy for your case list.
- Documents you upload. If you use AI features that let you summarize a document you upload (as opposed to a document Trellis already has in its corpus), you are handing that document to the vendor's servers and, potentially, to any subprocessor model provider.
- Alerts and saved matters. These are effectively a list of what you are watching and, by inference, who you represent or are targeting.
For each of these, you should read Trellis's privacy policy and terms carefully, and ask specifically: (a) is my query and document data used to train models, and can I opt out; (b) what is the retention period and can it be shortened contractually; (c) which third-party model providers, if any, receive my prompts; and (d) is a DPA available. The public policy is the floor; enterprise contracts can and should raise it.
On the subpoena question — what could a subpoena served on the vendor reach — the honest answer is: whatever the vendor retains. That includes account records, query logs, saved matters and alerts, and uploaded documents for as long as they are stored. A subpoena to Westlaw or Lexis would face the same theoretical exposure; the practical difference is that state trial court queries can be more revealing about a live matter than a citator lookup. Judges pay attention to metadata in discovery disputes, as the Otter.ai class action (Brewer v. Otter.ai, 2025) illustrates in a different context. Treat your Trellis query history as vendor-visible, and don't put anything into a free-text AI prompt you would not want opposing counsel to see if the vendor were subpoenaed.
One useful anchor here is Judge Rakoff's reasoning in US v. Heppner (S.D.N.Y. Feb 2026), which held that a litigant's chats with a public AI platform were not privileged. That case does not directly govern lawyer-vendor arrangements — a proper engagement with an appropriate DPA is a different posture — but it is a reminder that pouring case narrative into a prompt without thinking through the vendor relationship is a bad habit.
Ethics-Opinion Fit (ABA 512)
ABA Formal Opinion 512, issued in July 2024, is the current center of gravity for generative AI ethics guidance. It emphasizes competence (Rule 1.1), confidentiality (Rule 1.6), communication with the client, and reasonable supervision of the tool's outputs. New York City Bar Formal Opinion 2025-6 (Dec 2025) pushes in a similar direction with more granular guidance on client disclosure and billing.
Trellis fits reasonably well against those duties, with caveats:
- Competence. Understanding the tool includes understanding that AI-generated summaries of dockets and orders can be wrong or subtly misleading. State trial court dockets are messy; entries can be re-classified, sealed, or amended. Treat AI summaries as a starting point and read the underlying filing before you rely on it in a brief or an advice letter.
- Confidentiality. The core research use — searching public dockets — is low-risk. The higher-risk moves are uploading a client's draft complaint for "analysis" or pasting sensitive facts into an AI prompt to ask what similar cases have settled for. Those interactions are governed by the vendor's terms and any DPA you negotiate.
- Supervision. Under Rule 5.1 and 5.3, partners are responsible for how associates and staff use these tools. A firm policy on what can and cannot be pasted into Trellis AI features is a very cheap risk-reduction step.
- Verification. Never cite a case, docket entry, or judge statistic from Trellis (or any AI-assisted tool) without confirming it against the underlying record. That is basic Mata v. Avianca hygiene.
None of this is unique to Trellis; it is the standard checklist for any AI-augmented research product. But it is worth working through explicitly before you roll the tool out firm-wide.
Where Trellis Is Genuinely Strong
Credit where it is due. Trellis's coverage of state trial courts is materially better than what the generalist legal research vendors have historically offered. For a California litigator, the ability to run a single query across Los Angeles, Orange, San Diego, Alameda, and San Francisco superior courts — and to actually pull the filings — is a real time saver. Insurance defense teams use it for exposure analysis. Plaintiffs' firms use it to identify defendants with a pattern of similar conduct. Business-development teams use docket alerts to find prospects.
The judge analytics feature, if used with appropriate skepticism about sample size and selection bias, can inform genuinely better strategic decisions — motion timing, venue analysis, and settlement posture. Just do not confuse a base-rate statistic with a prediction about your particular judge and your particular motion.
Pricing and Who It's For
Trellis does not publish a price list. Pricing is quote-based and depends on jurisdictional coverage, seat count, and feature bundle. The current pricing page routes to a demo request. Expect it to be a meaningful subscription line item — this is a professional-grade data product, not a consumer tool — and expect enterprise buyers to be able to negotiate the DPA and data-handling terms.
Who it is for:
- State-court litigators, especially in high-volume jurisdictions with weak native e-filing search.
- Insurance defense and personal injury firms doing systematic party and counsel research.
- Plaintiffs' class action shops looking for repeat-defendant patterns.
- General counsel evaluating outside counsel or watching adverse filings against the company.
Who it is not for: transactional lawyers who never see the inside of a courthouse, and small-firm generalists whose research is 90% appellate authority already covered by their existing Westlaw or Lexis subscription. Trellis is a complement to those tools, not a replacement.
Trellis pairs well alongside a dedicated meeting-capture tool for client intake and witness prep — Basil for Law runs entirely on-device and keeps that layer of the workflow out of any cloud vendor's logs — but they solve different problems and should be evaluated on their own merits.
Verdict
Trellis is a serious tool for lawyers who live in state trial courts. The underlying data is the real asset; the AI layer is useful but should be treated as an assistant, not an authority. Confidentiality-wise, the main discipline is to treat your queries and any uploaded documents as vendor-visible and to negotiate a DPA if you are at any scale.
| Pros | Cons |
|---|---|
| Best-in-class state trial court coverage and search | Query and alert metadata is a fair proxy for your case list; treat accordingly |
| Judge, party, and opposing counsel analytics that generalist tools lack | AI summaries need verification against underlying filings, always |
| Meaningful workflow gains for litigators, especially in California | Pricing is opaque and quote-based |
| Docket alerts are genuinely useful for BD and active-case tracking | Uploading client documents into AI features raises the confidentiality bar; read the DPA |
| Complements rather than duplicates existing Westlaw/Lexis stack | Limited value for pure transactional or appellate-only practices |
Bottom line: if you handle state court litigation at any volume, Trellis is worth a demo. Go in with your confidentiality questions written down, ask specifically about training use, retention, subprocessors, and DPA availability, and set an internal policy on what your team is and is not allowed to paste into AI prompts. Treated that way, it is a risk-reducing productivity gain rather than a new exposure.
This review is for information only and is not legal advice.
Frequently asked questions
Does Trellis replace Westlaw or Lexis?
No. Trellis is focused on state trial court records, dockets, and analytics — an area where the generalist legal research vendors have historically been thin. It complements Westlaw or Lexis for appellate authority, statutes, and secondary sources rather than replacing them. Most litigation shops that adopt Trellis keep their existing research subscription.
Is it safe to paste client information into Trellis AI features?
Treat any free-text prompt or document upload as vendor-visible until you have confirmed otherwise in writing. Read the current privacy policy and, at any firm scale, negotiate a data processing agreement covering training use, retention, and subprocessors. For high-sensitivity matters, restrict AI-feature use to public docket content and keep client-specific facts out of prompts.
How does Trellis fit with ABA Formal Opinion 512?
Opinion 512 emphasizes competence, confidentiality, communication, and supervision when using generative AI. Trellis fits reasonably well provided lawyers verify AI-generated summaries against the underlying filings, understand where their query and document data goes, and set a firm policy on acceptable use. The tool does not, by itself, discharge those duties — the lawyer does.
Can a subpoena to Trellis reach my search history?
In principle, yes — a subpoena to any vendor can reach whatever the vendor retains, subject to the vendor's terms and any DPA, plus applicable law. That includes account records, query logs, alerts, and uploaded documents. This is not unique to Trellis; it applies to any cloud-based research or AI tool. The mitigation is to minimize what you type or upload and to negotiate retention terms in your contract.
How reliable are Trellis's judge analytics?
They are a useful starting point but should not be treated as predictions. Base-rate statistics about how a judge has ruled on categories of motions are affected by sample size, selection bias in the cases that reach that judge, and changes in the judge's docket over time. Use the numbers to inform strategy, not to make guarantees to clients.
What does Trellis cost?
Trellis does not publish a public price list; pricing is quote-based and depends on jurisdictional coverage, seats, and features. Expect it to be a meaningful subscription line item appropriate to a professional-grade data product. Check the vendor's pricing page for current details and request a scoped quote based on your practice footprint.
Meeting notes with no server to subpoena
Basil transcribes and summarizes entirely on-device — privilege-safe by architecture. See Basil for Law →
This review is for information only and is not legal advice.