Professional legal research, made better by those it serves.
Agentically researched from the public record, and free for anyone to search from your AI assistant, with no daily limit. Platforms can build it into their own products.
Prior art, a claim, a rule, an examiner. Free, no account. We never keep your questions.
Search the legal record
Legal questionsFree, no account
SearchesIP & prior art patents, prosecution, examination guidanceCase law · Tax · Contracts · Litigation · Corporate soon
Legal AI is only as good as what it retrieves.
Sourced
Benchmarked to the professional record.
Patents, court dockets and agency filings come from public sources, then each area is measured against what professionals already decided.
- Patents
- examiner citations
- PTAB
- Board decisions
- Case law
- how courts treat earlier cases
Researched
Agentically researched.
Every method is scored against that professional benchmark before anyone relies on it.
- Split
- into what the question turns on
- Searched
- many ways, across every garden
- Scored
- against the benchmark, in the open
Enriched
Community enriched.
Searching is free. When you or your agent spots a wrong record or a missing link, report it.
- Checked
- by our experts and review agents
- Shared
- the fix reaches everyone
- Credited
- accepted reports earn credits
What we are
Built in the open, layer by layer.
You search it for free and can enrich it along the way.
We fix it, keep it current and publish how well it works.
Raw data
Patents · Prosecution · Guidance · Case law
Research
Research · Benchmarks
Representation
Embeddings · Knowledge graph
Refinement
Rerank · Cleaning
Your integrationyours
Claude · ChatGPT · Gemini · MCP
What does your legal AI base its trust on?
IP is free to search today, with prosecution and examination guidance inside it; the rest are on the roadmap.
Measured on PTAB-invalidated claims
93%
of claims the Patent Board invalidated had the Board's prior art, or art as strong, in our top 10.
54 had it4 did not
54 of 58 held-out claims whose Board reference is in the index; "as strong" judged by a model. On the 97,913-query public holdout, v1 finds 45.3% of examiner-cited art in the top 100 and 74.2% in the top 1,000, trained on 609,723 examiner citations (fingerprint 6fd3de94f3b9). Full benchmarks→
What IP & prior art holds
10.5M+ US patents
Patents and applications
10.5M+ US grants and published applications, parsed to claims and specification.
Citations
48.5M examiner and applicant references, followed forward and backward.
Patent space
Every filing embedded and placed, so neighbours and gaps are coordinates rather than keywords.
Claim structure
Antecedent basis, and which passage of the specification supports each limitation.
What Prosecution holds
10.3K examiners, 8.5M CPC classifications
CPC scheme
The full classification hierarchy, symbol to definition to children.
Classification
Main CPC symbol for 8.48M publications, resolvable from an application number.
Art units
Where applications carrying a symbol were routed, and what each unit handles.
Examiners
10,308 profiles, with how each one has read claim language before.
What Examination guidance holds
MPEP and EPO Guidelines, 3.5K sections
MPEP
Sections parsed to the rule each one turns on, cross-referenced both directions.
Form paragraphs
The text an examiner pastes into an office action, keyed to the rejection it carries.
EPO Guidelines
The European counterpart, parsed the same way so the two can be read against each other.
Design codes
The design classification scheme, symbol to definition.
Phase 2
Case law is on the roadmap.
Precedential and persuasive decisions, parsed to holdings and the authorities each one turns on, traceable back to the reporter.
Next on the pipeline. The same pipeline that indexed patents, pointed at a new corpus.
Phase 2
Tax is on the roadmap.
The Internal Revenue Code, Treasury regulations and IRS guidance, parsed to sections and linked to the rulings that interpret them.
Next on the pipeline. The same pipeline that indexed patents, pointed at a new corpus.
Phase 3
Contracts is on the roadmap.
Executed agreements broken to clauses and defined terms, so a call retrieves the operative language instead of a whole PDF.
On the roadmap. The same pipeline that indexed patents, pointed at a new corpus.
Phase 3
Litigation is on the roadmap.
Dockets and filings resolved to the claim that was litigated and how it landed, linked to the parties and the court.
On the roadmap. The same pipeline that indexed patents, pointed at a new corpus.
Phase 4
Corporate is on the roadmap.
Corporate filings with assignees and registrants resolved to real entities, so ownership chains connect across the record.
On the roadmap. The same pipeline that indexed patents, pointed at a new corpus.
Two ways in
Use it where you already work.
Free · For everyone
MCP server
Connect your assistant and search, free and with no daily limit.
Any prior art for this claim? It covers encrypting ledger state before consensus…
The closest art is US10958423B2, which discloses encrypting ledger state prior to consensus in claim 1. The matched passage: "…the state tree is sealed before votes are…"
- Claude, Copilot, ChatGPT, or any MCP-compatible client.
- Spot a wrong or missing record? Report it from the same chat. A person reviews it, and an accepted fix reaches everyone in the next build.
- Results cite the record, down to the passage behind each hit.
For platforms
Data API
Building a product? Put the same search inside it.
curl https://mcp.antecedint.com/api/v1/prior-art \ -H "Authorization: Bearer $ANTECEDINT_KEY" \ -d '{ "claims": "1. A method of…", "useReranking": true }'
// 200 { "success": true, "data": { "references": [ { "publicationNumber": "US10958423B2", "score": 0.91, "title": "Ledger state encryption…" } ] } }
- Legal records behind one REST endpoint, each parsed to its working parts: 10.5M+ US patents today, with prosecution and examination guidance beside them.
- A search tuned to the question you ask of each corpus. Every hit carries the passage it matched on.
- Free to start on Evaluate; Launch is $200 a month, metered in credits on your own key.
Questions
Common questions.
What is Antecedint?
Antecedint is open legal context: the public legal record, organized area by area in the shape each area's questions need, a ranked index for prior art in patents, a web of citations for case law, and the same care for tax and the areas after it. Anyone can search it for free from an AI assistant over MCP, and platforms build on it through a REST API. Every result comes back with the source passage it rests on. The people who use it, and their AI agents, report wrong or missing records; our experts and review agents check each report, and accepted fixes go into the next build for everyone.
How is it free, and what happens to my questions?
Search and context are free, and your questions stay yours: we do not keep them, log their text or train on them. What helps us is feedback on the public record. When you or your agent reports a wrong record or a missing connection, our experts and review agents check it, and an accepted correction goes into the data for everyone and earns you credits for deep search, tools and drafting. We keep it free by selling integrations to companies that build the same data into their own products. The search box on this site is a demo, which plans each search through Google's Gemini API without keeping it; the main way to use Antecedint is from the assistant you already use.
How is it accessed?
Two ways, on one key. An MCP server at https://mcp.antecedint.com/api/v1/mcp that Claude, ChatGPT, Copilot and any MCP-compatible client can add, and a REST API at mcp.antecedint.com with bearer authentication for platforms that build on it. Every MCP tool calls the same REST endpoints, so rate limits and metering are counted once.
What does it cost?
Search is free. Over MCP on your own key there is no daily search limit; the key's per-minute ceiling still applies. Credits pay only for deep search, the rerank, the tools and drafting. Every free key gets 500 credits a month with a verified email and no card, then 20 credits a day until the month resets, and accepted data reports through the MCP Feedback program earn more. Platforms on the API start free on Evaluate; Launch is $200 a month for 2,500 credits a month. Launch keeps serving to twice its monthly allowance and flags the overage. Nothing is charged automatically, and use past the allowance is forgiven when the month resets. Enterprise is an annual agreement for custom integrations and terms, with volume set by the agreement and every new corpus included as it ships.
How accurate is it, and how is that measured?
On a pinned holdout of 97,913 queries, scored on whether the art an examiner actually cited appears in the top 100 results, Antecedint retrieves 45.3%. PatentSBERTa, a patent-domain fine-tune, retrieves 21.4% on the identical harness, and BM25 retrieves 24.3%. The harness, the split and the negative results are published, and anyone can rerun the benchmarks on one GPU in an afternoon.
Which corpora are live?
Three ship today. IP and prior art: 10.5M+ US patents and published applications, 48.5M examiner and applicant citations, and the claim-structure tools. Prosecution: the CPC scheme, the main classification for 8.48M publications, art-unit profiles and 10,308 examiner profiles. Examination guidance: the MPEP and the EPO Guidelines for Examination parsed to sections, with the form paragraphs an examiner pastes into an office action. Case law, from CourtListener and GovInfo, and codified law, from OLRC and eCFR, are being built; contracts are on the roadmap. None of them carries numbers yet.
Does Antecedint sell an AI agent or a chatbot?
No. Antecedint does not sell an application that competes with the people who build on it. A person searches it from their own assistant; a platform, firm or assistant calls it from its own product and owns the interface, the workflow and the client relationship. Antecedint keeps the corpus current, fixes what users report, and publishes how well the retrieval works.
Antecedent: what came before. The record a claim has to answer to, and the thing a model has to have seen.
Find the clearing.
Bring an idea and search for free. Leave with the closest art, and the receipts.