Trellis

What it is

Trellis is a research and analytics platform for US state trial courts — the dockets, motions, rulings and verdicts that never become published appellate opinions and therefore never show up properly in Westlaw or Lexis case-law search. It was founded by Nicole Clark, a former litigator, and claims the largest state trial court dataset in the country: 2.5 billion+ records from 3,500+ courts across 2,500 counties in 45 states. Litigators use it to answer questions case law can’t: how this judge has ruled on summary judgment, what opposing counsel files and wins in this county, what comparable premises-liability verdicts actually paid.

The closest comparison is LexisNexis’s Lex Machina, not a case-law tool. Lex Machina built its reputation on federal analytics and added state courts later; Trellis started in the state trial courts and has stayed there.

2026 is when it became a data source for AI assistants rather than a destination. Trellis launched an MCP connector for Claude on 12 May 2026, the same day Anthropic launched Claude for Legal. On 28 July 2026 it added a ChatGPT plugin and Trellis Chat, an in-product agent that plans multi-step research across the trial court data and returns a sourced answer. An API sits underneath all of it.

Why it shows up in Legal Ops stacks

• It covers the part of litigation where most of the work happens. Trellis’s own figure is that 98% of US civil litigation runs through state trial courts. Westlaw and Lexis are built around reported decisions; a trial-level ruling on a motion in limine in a county court usually isn’t one.
• Judge and opposing-counsel analytics at county level. Motion grant rates by judge, attorney track records, party litigation histories and expert witness records — the inputs to a venue or strategy memo that otherwise come from asking colleagues who’ve appeared there.
• It is the trial-court connector in a Claude for Legal deployment. Claude does the reasoning and drafting; Trellis supplies the county-level record. Our Claude for Legal page names it alongside CourtListener and Midpage as the research connectors that stop citations coming back unverified. See best MCP tools for legal ops.
• Alerts on judges, parties and topics. New filings against a client, a competitor or a repeat opposing party arrive by email — a cheap early-warning feed for an in-house litigation team.

Pricing reality

Trellis publishes its self-serve tiers in its help centre (the pricing page itself sits behind a free sign-up):

Two things decide what you actually pay. First, every self-serve plan is limited to one state. A firm litigating in California and Texas is already in the custom tier. Second, everything is metered in Content Views — searches, alerts, judge pages, case updates and each Trellis Chat question, which uses one view or more for broad multi-part questions. The Claude and ChatGPT connectors are included with any subscription but draw on the same allowance. Monthly-plan add-ons run $23 for 20 extra views up to $100 for 100.

The realistic number for a solo or boutique litigator who wants the reason most people buy it — judge analytics — is about $2,000 a year for one state. Multi-state firm licences are quote-only; budget them as an enterprise data contract, not a seat.

Best for

Litigators and litigation support teams at plaintiff firms, insurance-defence firms and in-house legal departments whose matters sit in state trial courts, who need judge, motion and opposing-counsel intelligence for a specific venue, and who want that data inside Claude or ChatGPT rather than in another research tab.

Not for: federal-court practices, appellate work, transactional teams, or anyone who needs secondary sources, treatises or statutory research — Trellis has none of that depth. It complements a primary research platform; it does not replace one.

Alternatives and when to pick them

• Thomson Reuters CoCounsel with Westlaw Litigation Analytics — the market-share incumbent. Pick it when the firm already pays for Westlaw and your state-court questions are occasional; the analytics come with the contract you have.
• LexisNexis Protégé with Lex Machina — the other incumbent. Lex Machina covers 1,300+ state courts in 34 states and DC with deeper federal analytics. Pick it when the practice splits federal and state and you want one analytics vendor for both.
• Midpage — the AI-native entrant most often paired with Claude for litigation research. Pick it when the job is case law and citation checking inside Claude at $100/user/month. It answers “what’s the law”; Trellis answers “what does this judge do with it.” Many Claude for Legal deployments want both.
• UniCourt or Docket Alarm — pick one of these when the need is docket tracking and court-data pipelines across state and federal courts, rather than judge analytics at the county level.
• CourtListener — free, from Free Law Project, with federal dockets via RECAP and an MCP server. Pick it when the budget is zero and the matters are federal.

Watch-outs

• Coverage is uneven by state and county. 45 states is not 50, and county-level depth varies with what each court publishes. Guard: before buying, look up five of your closed matters and three judges you’ve appeared before in the free account, and confirm the dockets and rulings are actually there.
• An agent burns Content Views faster than a person does. One broad question in Claude or Trellis Chat can consume several views, and the Research tiers carry 900 a year — about 75 a month. Guard: run a two-week trial at real agent volume, track views consumed per matter, and size the plan or negotiate a firm-tier allowance from that number.
• Analytics are only as good as the docket’s coding. Grant rates and outcomes are derived from how county clerks record filings, and Trellis publishes no independent accuracy audit. Guard: pull the underlying rulings behind any statistic that goes into a client memo or a venue decision, and cite the rulings, not the percentage.
• Third-party reviews flag slow support. A March 2026 review noted customer-support response-time complaints. Guard: on a firm licence, get a named account contact and a response-time commitment in the order form.

For the category around it, see best legal AI research tools and grounding vs hallucination in legal AI.