Human review
Every proposed value carries a confidence score. A person accepts, corrects or rejects low-confidence values; some cases always need a person.
Use case · Government, defense and legal
AI proposes. A person confirms. Every step is on the record.
For legal teams and government and defense-adjacent work: contracts, NDAs and legal agreements, including JAG-related legal work. Built around what these teams screen for: human review, citation to the passage, need-to-know access and a record that can be reproduced.
In developmentNothing is in production yet.
Decision record · one documentSample data
- 1Proposed · Adduce09:14:02Classify Teaming agreement, sample program as 2Counselp.1 headerp.7 §9Conf. 0.88
- 2Corrected · A. Chen09:31:47Changed to 3Restricted. Note: “Names program staff.”
- 3Recorded · audit trailIn designWho, what, when and the evidence shown, kept in a tamper-evident record.
The documents
Agreements that have to hold up.
NDAs
Terms, notice periods, survival and governing law, with the clause behind each.
Master services agreements
Liability caps, termination and renewal, compared across a repository.
Legal agreements
Lease amendments, teaming agreements and invoices, split and filed with their own access.
JAG-related legal work
Document work where need-to-know and a defensible record come first.
The controls
What a reviewer can check for themselves.
Citation to the passage
Each answer names the document, page and section it relied on. Answers stay marked “not verified” until someone checks.
Need-to-know access
Access per repository and per document, with clearance tiers your organization defines, enforced inside search.
Reproducible answers
The same question from people with the same access gives the same answer. Tested as a guarantee.
Audit trail
Viewing and sensitive actions recorded in a tamper-evident trail, with thresholds you set by tier.
Contradiction finding
Clauses that conflict across documents, shown to each person only as far as their access allows.
Where things stand
The plain version.
| Question | Answer today | Status |
|---|---|---|
| Where does it run? | Adduce-run, on Microsoft Azure, with each customer isolated from the others. | In development |
| Which models read our documents? | Open-weight models hosted on Adduce infrastructure. No public AI API sees document text. | In development |
| Can it run in our environment? | Not today. On-premises, in-boundary and government-cloud deployment are not offered. | Not offered |
| A government edition? | A separate government edition is a planned direction, not a commitment. | Planned |
| Certifications and authorizations? | We hold none today. SOC 2 Type II and HITRUST CSF are planned targets. | Planned |
| Is anyone using it? | Not in production yet. | In development |
Government and defense programs
Talk to us about your program.
Choose “Government or defense” on the contact form, or email info@adducelabs.com, so your note reaches the right person.