The Verified Interface
For Legal AI
Real law in, every citation checked out. Metered by the hit, no seats.
Keep the model you already use. Before it answers, we hand it what the law actually says; after it answers, every citation is checked against our library. Anything unverifiable is removed, and the answer says so.
See the product
Real screenshots of the running product, not drawings of it.
Check a briefCheck a brief
Six citations pulled out of a pasted brief, five found, one reported as unverified.
Review a contractContract review
A clause read against the state's verified rules, with the authority behind each one.
Open timekeepingTimekeeping
Draft entries from the work you already did here, reviewed before you bill.
Free for lawyers
The product demonstrating itself, and tools built from data you already generate here. No seats; what is metered is on the rate card.
Check a brief
no accountPaste a draft; every case citation in it is checked for existence against 18,123,788 citations across 989 reporters. Unverified is said as unverified, never as fabricated. Nothing stored.
Check a briefTimekeeping
sign inDraft time entries from the work you already did here, reviewed against each client's billing guidelines, exported to CSV, LEDES or Clio. A floor on your time, not an estimate.
Open timekeepingCase-file scrambler
sign inPseudonymized on our hardware before any frontier model sees it; names come back in the answer; the document restores byte-exact. In the browser or by API.
Open the scrambler
Featured RAGs
The corpora behind the grounding step, callable on their own: 615,391 Texas decisions, 122,681 statute sections, 3,860 verbatim rule sections and 18,123,788 reporter citations.
- Texas law index
Hybrid search over 615,391 Texas decisions, 122,681 Texas statute sections, 45,660 U.S. Supreme Court opinion documents and 370 county local rules, on our own hardware.
615,391 decisions122,681 statute sections370 local rules - Verbatim rules
3,860 sections of court rules and statutes held word for word: Texas civil, evidence and appellate rules, the federal FRCP, FRE and FRAP, Texas codes and U.S. Code titles. (107 more sections, NY CPLR and CA CCP, are held in the download but excluded from every jurisdiction's search.)
3,860 sectionsBM25 by short citeJSON download - Citation table
18,123,788 reporter citations across 989 reporters, built from the Free Law Project bulk export. A hit proves a citation exists; a miss proves nothing, and the endpoint says so.
18.1M citations989 reportersSQLite download
Contract review
betaSend a contract and a state. Each clause comes back with the state's verified rule, its authority, and whether the clause appears to trip the rule's known trap.
- Verified rules
- 602
- States
- 28
- Texas families
- 21
- The model labels clauses and reads traps. Every rule and citation is copied from a verified store.
- A clause family with no verified rule returns
no_verified_rule. - Contracts are not stored. A research aid for attorneys, not legal advice.
Verified rules by state
By clause family →- Texas26
- California24
- New Jersey23
- Alabama22
- Connecticut22
- Iowa22
- Illinois22
- Kansas22
- Maine22
- Minnesota22
- Montana22
- North Carolina22
- Nebraska22
- New York22
- Ohio22
- Oregon22
- Rhode Island22
- Utah22
- Virginia22
- Wisconsin22
- Delaware21
- Florida21
- Washington21
- Idaho20
- North Dakota20
- Alaska18
- Louisiana17
- South Dakota17
Every other state is reviewed for clauses and returns no verified rule.
Docket Scrambler
send the case, not the clientA case file is pseudonymized on our hardware before any frontier model sees it, with every person, organisation and identifier replaced by a stable per-matter placeholder, and the names come back in the answer. The document restores byte-exact. The mapping is encrypted per matter and never returned by any route.
- 0 of 1,984
- entities leaked, 40 real dockets
- 6,200/6,200
- byte-exact restores
- 0 of 3,994
- citations harmed
How DocketRouter works
One loop around the model you already use. The teal stages are ours; the reasoning is the enterprise model's.
- Ground
Your question, not the whole record, queries a fused legal index on our own hardware.
- 615,391 Texas decisions, 122,681 statute sections, 45,660 U.S. Supreme Court opinions, 370 county local rules
- 776 Texas rules of court and 228 federal rules held verbatim, quoted by short cite
- About a second per lookup
- Reason
The enterprise model you chose reads the retrieved authority and writes the answer.
- A small in-house model grounds and checks; it never writes the answer
- The answer stays the enterprise model's own
- DeepSeek is enabled today; keys can be set to in-house models only
- Verify and re-answer
Every citation in the answer is checked, up to 80 per answer, against our index first.
- If any fail, the model is told which were confirmed, unverifiable or nonexistent and asked to re-answer, up to two rounds
- Whatever is still unverifiable is removed mechanically and the answer says so
- Nothing unverified ships as if verified
- Disclose
The response carries what was retrieved, what was checked, what was removed, and what it cost.
- Removed citations are listed in the verification report
- A degraded flag when retrieval was unavailable, never a silent fallback
- A quarantine report for every client document read
What every response carries
Standard OpenAI-compatible completion, plus a docketrouter object. Same fields whether the answer was clean or had citations removed.
- sources
- the rules and opinions that were retrieved and put in front of the model
- verification
- every citation in the answer, its status, and the list of any that were removed
- injection
- clean, suspicious or hostile, per client document read
- degraded
- set when the index was unreachable or nothing matched, so an ungrounded answer is never read as grounded
- usage.cost
- exact billed cost per request, with the upstream cost beside it in your usage export so the markup is auditable
Your data
Legal records are routinely written by the other side. The layer is built on that assumption.
Uploaded documents are treated as evidence, never as instructions: fenced, hidden-character smuggling stripped, and scanned for embedded instructions. The scanner is tuned against real legal prose so ordinary filings do not trip it. The verdict for each document is in the response.
A key can be set to a policy where only in-house models are allowed. Under that policy the prompt never leaves the box, and routing to providers that retain or train on data is denied.
Each key carries a hard spend cap that is enforced upstream. Spend can never exceed the cap.
A matter holds a message history and its attached case files, encrypted per client, and narrows retrieval to that matter's documents.
- 1Gemini 3.7 Flashraw58.6%with DocketRouter64.2%by Google · $0.94/M in · 384 items
- 2Solar Pro 4raw39.2%with DocketRouter50.7%by upstage · $0.11/M in · 909 items
- 3DeepSeek V4 Flash 0423raw42.5%with DocketRouter50.2%by DeepSeek · $0.01/M in · 999 items
- 4gpt-oss-120braw24.5%with DocketRouter38.0%by OpenAI · $0.05/M in · 960 items
- 5gpt-oss-20braw2.2%with DocketRouter25.8%by OpenAI · $0.02/M in · 659 items