Skip to content

Trust center

Trust should be inspectable. 

Vaultr is designed so its privacy, security, evidence and governance claims can be examined rather than merely accepted. This page is the register: every claim, the scope it holds within, the evidence behind it, and an honest status where the evidence is still being built.

Claims apply to the current public build · last reviewed 15 September 2026 · runtime evidence: Not independently published

01

Trust principles

01

Local by design

Local Mode is the default. The architecture is built so matter processing works with no vendor in the path.

02

Evidence over assertion

Findings carry the passage they came from. Claims on this site carry the page that supports them.

03

Human-controlled actions

The application proposes; counsel decides. No change is applied without a recorded human decision.

04

Inspectable architecture

The source is public under AGPL-3.0, and the tests that check the boundary are published for you to run.

02

The data boundary

In Local Mode, every stage of a matter task is designed to stay inside the Vaultr boundary — no vendor, no upload dialog, no silent fallback. The visualization below shows the flow, and the toggle shows exactly where it changes.

Vaultr local boundary
  1. Matter
  2. Parsing
  3. Local inference
  4. Evidence
  5. Human review
  6. Output

In Local Mode the six stages above stay inside the boundary: matter content, prompts, inference and outputs remain in the local Vaultr environment. Network exceptions are the ones you start on purpose — updates and model downloads.

03½

Run the boundary test

One-click test packetmacOS · run on a dedicated test machine

Download the helper, freeze the release, run the fixed workflow, and inspect the evidence it leaves. This is a scoped way to test the boundary yourself — not a universal zero-egress certificate.

Reviewed
15 September 2026
Scope
Current public build · Local Mode · one tested macOS configuration
Runtime result
Not independently published
Boundary
A test result must be tied to the release, hardware and workload that produced it.
03

The proof register

Each row is a claim, held to a stated scope, with the evidence that exists today and a status that does not overstate it. "Verified", "certified" and "audited" are not used on this page because that work has not been done yet — and the restraint is deliberate: it is what makes the register worth checking.

9 claims · statuses change when the evidence does, not when the marketing does

04

Open-source licence compliance

OpenChain — ISO/IEC 5230Self-certified · September 2026

ISO/IEC 5230 concerns open-source licence compliance Open-source licence compliance: the processes a programme uses to manage licence obligations across the software it ships. It is a governance standard, not an information-security certification, and nothing on this site describes it as one.

OpenChain Conformant badge — self-certified

OpenChain Conformance — self-certification · September 2026

Vaultr's open-source licence compliance programme conforms to OpenChain ISO/IEC 5230 by self-certification. Self-certification is an officially supported conformance path under the OpenChain specification. It is a statement about our own programme processes — it is not an independent audit, and it is not an information-security certification.

Licence inventory
AGPL-3.0 for Vaultr itself, with upstream project licences documented in the repository.
Obligation tracking
Copyleft scope is stated in the security brief: internal firm use, with source obligations if third parties interact with a deployed instance.
Provenance
The public repository carries the licence text, the fork's upstream attribution and the review history.

Self-certification is an officially supported conformance path under the OpenChain specification — and we state it as exactly that. We publish the scope of what it covers rather than letting a badge imply more than it does; if we commission independent assurance, its result will appear here.

05

Responsible claims

The house rule

Security claims should have scope.

Benchmarks should have configurations.

Evidence should have provenance.

Where this site states a technical claim, it links to the methodology, the code or the demonstration that supports it. Where evidence does not yet exist, the register says "pending" rather than dressing the gap up as a guarantee.

06

Security contact

Reporting a security issue

Contact sabarinath.1805@gmail.com. Please do not include matter data or credentials in reports.

Open a security advisory on the public repositoryPublished: /.well-known/security.txt

Check it yourself

The strongest report is a reproduced one: run the boundary test, read the source, walk the register. Start with the security brief's verification procedure.