Module 04 / zero-egress AI
Document Q&A — Zero-Egress AI
This page mocks the platform's retrieval-augmented document assistant. No model is called and no network request is made — the preset questions below replay curated answers to demonstrate the interaction contract of a zero-egress design.
The contract has three parts. The corpus is curated: documents are indexed by an administrator, never uploaded by end users. Inference is local: the model and the index live next to the data, so nothing leaves the environment. And every answer carries page-level citations, linking each statement back to the exact source passage, so a reviewer can verify it in one click.
Ask the document
Pick one of the preset questions; the answer appears with its citations.
Fictional protocol written for this demo. Indexed locally; no content leaves the server.
In production this runs on a local LLM with a locally built index — zero external API calls. This mock replays curated answers to show the citation contract.
Source document
The indexed pages of the corpus. Passages cited by the current answer are highlighted.
The document, questions, answers and citations on this page are synthetic, generated by a seeded script (tools/platform-demo/generate-data.mjs) and embedded at build time. No model is called, no network calls.