Frequently Asked Questions

Packs only answer from retrieved passages that clear an evidence threshold. Below it, the server returns a structured refusal instead of a completion. Every answer carries its source citation and chunk key, so you or your agent can check it against the source.

A frontier model answers your patent question either way, fluently, whether or not it actually knows. You cannot tell those two answers apart. A DEX pack answers only from a vetted expert source and hands you the citation. When the source is not there, it refuses instead of improvising. Use the model for reasoning; use a pack for facts you have to get right.

Search brings back whatever the web says, and your agent still has to decide what to trust. A pack is a single vetted corpus for one domain, with an evidence threshold and a refusal mode. It is narrower on purpose: it covers less and is wrong less.

Each answer runs retrieval over the corpus, an evidence check, and a citation pass before anything comes back. Refusals return faster, usually four to five seconds, and cost nothing. The tradeoff is the point: slower than a raw completion, but checkable.

No. It is retrieval over the public-record MPEP with citations: examiner procedure, not attorney advice.

One expert's knowledge base, versioned and served as an MCP endpoint your agent can call. It answers with citations, or it refuses.

All three launch packs were curated by the founder from public primary sources. No outside expert, and no source organization, built or endorsed them. Founding creators will be the first outside experts on DEX.

Tax (IRS publications), Immigration (USCIS policy manual), Aviation (FAA regulations), Securities (SEC rules and guidance), Workplace safety (OSHA standards), and Environmental (EPA regulations). Every future pack keeps the same clean-room standard as the MPEP pack: a public primary source is required.

You could. You'd spend weeks curating sources, tuning chunking and retrieval, and deciding when to refuse, and then you'd maintain all of it. A pack is that work already done, for two cents a question.

The console replays real recorded sessions: questions, answers, citations, latencies, and prices exactly as the live packs returned them on 2026-07-07. No sign-in is needed to inspect those. To run your own queries you need a beta API key. Join the list below.

Join the beta below. When your batch is onboarded, your invite email carries a beta key and your pack endpoint. Claude Code is the tested client: you add the pack in one line, claude mcp add dex-patent, followed by the endpoint issued with your key. There is no public URL yet; endpoints are issued per beta batch.

We onboard in small batches. When your turn comes, your invite email has your beta key, your endpoint, and a two-line quickstart. Until then everything here is inspectable without an account: the console replays real recorded sessions, and the full transcripts are public.

Founding creators lock in an 80% revenue share, in writing, before the marketplace opens. $0 to list. You earn when your pack answers; earnings start when billing turns on.

Today, DEX is three expert packs your agent can call. The goal is a marketplace where people who know a field cold document the judgment their profession never wrote down, and get paid every time an agent draws on it. We think most professionals will eventually direct a fleet of agents, and those agents will need what real practitioners know. We are building that supply now, one pack at a time, with receipts.

Your agent's queries go to the pack server, which retrieves from the corpus and generates the answer through Anthropic's API. We log queries, retrieved chunks, and answers to debug retrieval and run evals. We don't use your data to train models and we don't sell or share it.