A legal research memo whose every citation is retrieved from real caselaw, checked, and Shepardized, so nothing on the page is a case an AI invented.
CODEX searches the actual caselaw and statutes first, synthesizes only from what it found, then verifies every citation the draft leans on before it ever hands you a memo.
The problem
The whole reason legal AI is scary is the fabricated case. A model writes a fluent, confident memo, drops in a citation that looks perfect, and it turns out the case does not exist (or it exists but was overruled two years ago). That used to be an embarrassment. Now it is a sanction. Judges have started issuing standing orders that require you to certify every AI-assisted citation, and the fines have followed.
So counsel are stuck in the gap the whole industry is stuck in: almost everyone is using AI for research, and almost nobody trusts it enough to actually rely on the output. The tool writes first and leaves you to find out later which cites hold.
How it works
CODEX reverses the order. It grounds first, then writes, then verifies, and it shows its work.
- Search real authority. It queries actual caselaw and statutes for your question and jurisdiction, and builds a research record out of what it retrieved (not out of what a model half-remembers).
- Synthesize from what it found. It drafts the memo narrative grounded strictly in the retrieved authorities, so the argument rests on real cases and statutes rather than on the model's confidence.
- Verify every citation. It parses each citation the draft relied on, checks that it resolves to a real authority, and Shepardizes it to confirm it is still good law.
- Draft the memo. Only citations that were retrieved, checked, and confirmed good law make it into the final memo. A cite that cannot be verified, or that is no longer good law, surfaces in the memo instead of being quietly drafted around.
One honest boundary, stated up front: CODEX researches and drafts, it does not file anything to a court on your behalf. Filing is a separate step that you confirm, by hand, after you have read what it prepared.
What you get
- A research memo on your question, grounded in caselaw and statutes it actually retrieved.
- A verified-citation trail: for every cite in the memo, evidence that it is real and still good law.
- A flag on anything it could not stand behind, surfaced rather than hidden, so you know exactly where to look.
- A deliverable built to survive the one question the court now asks, which is whether you checked your cites.
Who it is for
This is for solo and small-firm litigators and associates who write memos and briefs and have to personally stand behind every citation, and for in-house and legal-ops teams trying to decide whether an AI research tool is trustworthy enough to use on real matters. If you want a fast ungrounded answer and you are fine not checking the cites, this is not that, and it is not trying to be.
Pricing
Pricing is still being validated, so treat these as starting points rather than promises.
- Per memo. A flat price for one grounded research memo on one question: search, synthesize, and return the memo with every citation checked and Shepardized. You review before you rely on it. Starting hypothesis, around 99 dollars.
- Attorney seat. A monthly per-attorney subscription for unmetered grounded memos, for litigators who research constantly. Starting hypothesis, around 249 dollars a month.
- Firm platform. A firm-wide tier with shared matter grounding and the verified-citation trail retained per matter. Priced by conversation.
Any step that would move a document outward, or take a payment, is fail-closed and waits for a human to confirm it.
Proof, and how we dogfood it
AYA runs its own legal exposure through this exact discipline. The founder has a live housing-court matter, and AYA's regulatory-verification work turns on the same question every day, which is whether a cited authority is real and still good law. That is precisely what this deliverable's search, check, and Shepardize spine is built to answer, so AYA's own legal research is a genuine in-house user of it.
The deliverable behind this page was not freshly minted for a launch. There was already a research-memo pattern in the library, and it was broken (its steps pointed at the wrong machinery entirely). Rather than build a shiny new one, we repaired it to compose capabilities AYA already had (search caselaw, search statutes, synthesize, parse, check, and Shepardize citations, then draft), and we verified each of those pieces existed before this page went up. Building it left the same kind of receipt the product is about.
Honest note
Here is the part most landing pages leave out. As of today, the memo flow is defined and structurally sound, and its steps are composed correctly, but it is not yet running against live legal authority. The grounding depends on connections to the legal-research services (the ones that actually return caselaw and Shepardize a cite), and those connections are not credentialed yet in our build. So no memo has been generated against real caselaw for anyone, including us.
What that means for you, concretely: this is a product we are building in the open toward the moment those connections are live, not one you can run against a real question this afternoon. The word grounded is the entire promise here, so we will not describe a memo as grounded in real authority until a memo has actually been produced and verified against live caselaw, and witnessed. If the approach is what you have been waiting for, ordering a memo now tells us to wire the grounding first.
*This page is a specification. The capability it describes is not built yet, and nothing here is a claim that it runs today.*