Straight answers for lawyers evaluating Wythe. For security and confidentiality due diligence, see the dedicated Security & Trust page.
On this page: What is Wythe? · What's in the corpus? · Why is the search better? · What does the AI workbench do? · How do you handle hallucinations? · Is my work confidential? · What does it cost to try? · Is this legal advice? · Only the FEC? · Who built this?
What is Wythe?
Wythe is a vertically specialized legal research and drafting platform for federal campaign finance law. We took the FEC's public record — enforcement matters, advisory opinions, administrative fines, audits, alternative dispute resolutions, statutes, regulations, and the court cases around them — cleaned it, organized it, and built two things on top: a boolean search application that works the way lawyers think, and an AI workbench that researches, outlines, and drafts on that record. The two are separate: the search stands on its own, and the workbench is an optional add-on.
What's in the corpus?
As of July 2026:
- 7,600+ enforcement matters (MURs) spanning 1975 to today, plus 1,000+ ADR matters, 4,400+ administrative fine cases, 2,000+ advisory opinions, 1,000+ audits, and the federal statutes and regulations they apply.
- 265,000+ documents, virtually all full-text searchable — more than eight gigabytes of text.
- 184,000+ of those documents were recovered by splitting the FEC's monolithic scanned case files into individually labeled documents.
- 1.9 million+ citations extracted from the record and cross-linked into citing references.
An honest caveat: a small number of files are corrupted at the source — damaged uploads on the FEC's own site — so we don't claim literally every page of the record. Where something is missing or unreadable, we say so rather than pretend.
Why is the search better than the FEC's website?
- Boolean syntax lawyers already know. Root expansion (
contribut!), exact phrases,&,%, and proximity connectors (/s,/p,/n, ordered+n) — see the full syntax guide. - Split documents.The FEC posts many older case files as single enormous scans — hundreds of pages, many documents, one file. We split them into labeled documents (complaint, response, General Counsel's report, conciliation agreement…) so results point at the thing you actually want to read and cite.
- Snippets that land on the responsive page. Results show highlighted text snippets; clicking one opens the document at the exact PDF page with your terms highlighted.
- Citing references.For any statute, regulation, court case, or MUR, see what cites it across the record. The FEC doesn't offer a citator for its own materials; we built one.
What does the AI workbench do?
The workbench is a matter-based drafting environment. You pick a document type — memos, opposition/opening/reply briefs, complaints, complaint responses, letters, AO requests and comments, rulemaking petitions and comments, outlines, cite checks — upload your background documents, optionally give it an example to follow, and it researches, outlines, drafts, red-teams, and cite-checks on top of the corpus, court records, and (if you enable it) the public web.
It runs in two modes. In Plan mode it stops and shows you its outline first — you highlight text and leave inline comments, exactly as you would for a junior associate, and it revises before drafting. In Automode it runs straight through and marks assumptions where you didn't supply facts. Either way you can comment on and revise the finished draft, then export it to Word.
How does Wythe deal with AI hallucinations?
Language models work probabilistically, so they can invent citations — you've read the horror stories about lawyers filing briefs with fabricated cases from general-purpose tools. Wythe is built to eliminate or flag potential hallucinations by layering deterministic code around narrow AI prompts:
- Existence is checked by code, not by the model. Wherever a draft cites a case or an FEC document, hard-coded, non-AI checks confirm the cited authority actually exists in our corpus or in court databases.
- Characterization is checked narrowly.After a cite is confirmed real, we pull the document's text and send it to the model with just the part of the draft that relies on it, asking one narrow question: does the draft accurately characterize this authority?
- Uncertainty is flagged, not hidden.If the check can't be confident, the passage is flagged for human review — and those flags travel into the exported Word file as comments, so review happens where you already work.
No tool removes the professional obligation to review your own filing. Wythe is built so that review is fast and the failure modes are surfaced instead of buried.
Is my data used to train AI? Is my work confidential?
No, and yes. We do not train models on your content, and our AI providers are contractually barred from training on it either. Matter content is scoped to your organization by database-enforced row-level security, search queries are never stored, everything is hosted in the United States and encrypted in transit and at rest, and we run no third-party analytics. The complete due-diligence answers — including subprocessors, retention windows, and an honest account of our SOC 2 roadmap — are on the Security & Trust page.
What does it cost to try?
- Search: free for 30 days. Self-serve signup, no credit card required.
- AI workbench: a small, metered fee.Model APIs cost real money per run, so we don't let the workbench rip for free on the internet — the trial fee mostly converts into usage credit you spend on real drafting runs.
We're in beta and still calibrating subscription pricing, so we don't publish it here — ask us at info@wytheintelligence.com for current terms. One performance note: the self-serve demo runs on deliberately economical infrastructure. Enterprise deployments move to dedicated infrastructure with a dedicated search layer, which we expect to be substantially faster — for search and for the workbench runs built on it.
Is this legal advice?
No. Wythe is a research and drafting tool for professionals. It doesn't create an attorney-client relationship, and its output is a starting point for your judgment, not a substitute for it. See the Terms of Service for the full disclaimers, including our OCR-accuracy disclosure.
Do you only cover the FEC?
Today, yes — federal campaign finance is the specialty we built first, because vertical depth is what makes both the search and the AI actually good. But the platform generalizes: state campaign finance agencies, FERC, the FTC, the FCC, the Department of the Interior, PHMSA — if your practice lives in a regulator's record, we can build this for it. Tell us what you need: info@wytheintelligence.com.
Who built this?
Wythe Intelligence was founded by a litigator. In the founder's words: “For over a decade, a significant part of my practice has been federal and state regulatory litigation. There are great tools for searching caselaw — and almost none for searching agency decisions and documents. My wife has long had this problem with FEC materials, so I built the search tool with and for her. I built the AI workbench on top because friends and colleagues who used other AI legal tools kept telling me they were disappointed — and I had ideas about how to build AI legal tools that work better for lawyers.”
Questions, comments, or a regulator you want covered: info@wytheintelligence.com.