AI and data transparency
DocOps is open about how it uses artificial intelligence, how people supervise it, where document data is stored and what we commit to on security, so institutions can use the software with confidence.
Updated 04 Oct 2026
- 01
Where DocOps uses AI
DocOps uses large language models to read scanned pages, extract document details, classify, extract legal relations, answer questions and draft. The default is Anthropic's Claude; Agentra can configure another model per institution, such as GLM. Deterministic checks such as date, number and validity comparisons run on your machine without AI.
- 02
People decide
By human-in-the-loop design, a newly ingested document stays pending until a staff member reviews, edits and confirms it. Only then is it stored in the repository and indexed. Legal relations below 0.85 confidence also wait for staff approval; AI drafts are edited by staff and signed through your institution's own process.
- 03
Citations and no guessing
Every Q&A answer cites the document number and article it relies on. When the repository lacks a needed document, the system says "Not found in the repository" instead of generating an answer.
- 04
Data security
Original documents live on the user's computer. The server only runs the AI steps and keeps no content after returning a result. Each institution accesses with its own key rather than personal accounts, and institutions' data is fully separated.
- 05
Institutional knowledge belongs to you
The document repository and the legal relationship graph are your organisation's memory; they belong to the institution and can be exported at any time.
- 06
Legal compliance
DocOps follows Decree 30/2020/ND-CP on clerical work (the format of administrative documents) and Decree 13/2023/ND-CP on personal data protection.
- 07
Content ownership
The institution keeps all rights to the documents it uploads and to the processed results. Agentra never uses institution data to train models.
- 08
Limits of responsibility
DocOps supports professional work; it does not replace it. AI-generated content must be checked by staff before use; the final decision and the responsibility for issuing a document rest with the institution.
- 09
Policy updates
This policy may change when new features or new regulations arrive. The update date is shown at the top of the page.
- 10
Contact
Questions about AI and data: info@agentra.io.vn.