Security & Data

Data boundaries, local computation, and AI-provider limits.

Vnuli is a local engineering computation layer for AI-assisted investigation. This page explains the current data boundary so you can decide what to connect and what to ask.

The boundary

local engineering data
-> local Vnuli deterministic computation
-> bounded tool result / evidence / provenance
-> AI client/provider
-> explanation and follow-up

Where raw data lives

Raw engineering source files remain in user-controlled local files and projects. Vnuli registers those sources in local project state and interprets supported layouts from the installed Python package.

Vnuli does not make unsupported source layouts true by guessing. Use Supported Data to check which layouts are qualified.

Where computation happens

Vnuli computation runs locally through the installed package. When an AI client invokes a Vnuli tool, the local vnuli serve MCP server calls Vnuli's deterministic engineering tools against the registered project data.

The AI is not the numerical authority. Vnuli performs the calculation and returns evidence the AI can explain.

What the AI client can receive

Vnuli returns tool results through MCP. Depending on the tool and request, returned information can include values, statistics, summaries, metadata, provenance, warnings, plots, or other bounded tool results.

The connected AI client and its provider may therefore receive the information Vnuli returns through MCP. Review your AI provider's own data, privacy, account, and retention policies separately.

Bounded evidence

Vnuli is designed to return the evidence needed for the investigation instead of treating raw datasets as prompt context. For example, a bounded statistics or time-history request returns the requested result and supporting provenance rather than asking the AI to inspect an entire source file directly.

Telemetry

Current implementation inspection found no Vnuli product telemetry system. Vnuli itself does not currently send product telemetry.

That is not the same as saying no information ever leaves your computer: when you connect an AI client, the tool results Vnuli returns through MCP may cross the AI-provider boundary described above.

What this page does not claim

• It does not claim Vnuli is certified against a named security standard.
• It does not claim enterprise-grade security or compliance.
• It does not claim all data always stays on one machine after you connect an AI client.
• It does not replace your AI provider's privacy and security documentation.

Next steps