Back to Home

How Lunes works

From document to defensible thinking

Lunes turns the structure inside your document into a Brain that can support evidence-aware chat, focused persona review, and changes you control.

01
Build the Brain
02
Ask with context
03
Review from many angles
04
Choose every change

01. The Brain

A graph built from your document

When you first open a document, Lunes identifies claims, concepts, evidence, and relationships. It verifies contradictions before they become findings and keeps source links back to the document.

The Brain is derived from your document. It can be rebuilt, and persisted graph artifacts are encrypted with the project key.

02. Chat

Answers grounded in relevant context

Each chat turn retrieves relevant claims and verified contradictions from the Brain. Lunes sends that bounded context, your question, and the necessary conversation history to the model route selected for the document.

Built-in AI uses Lunes servers. BYO, local, and custom routes go directly from your browser to the provider you configured.

03. Personas

One document, several expert lenses

Personas review the same document and Brain context from different professional perspectives. Their findings stream into one review, where challenges and disagreements remain visible instead of being flattened into a single voice.

Partial suggestions are encrypted and checkpointed during the run, so an interruption does not erase all completed work.

04. Apply suggestions

Nothing changes until you approve it

A suggestion is converted into a staged document diff. You inspect the proposed wording in context, accept or reject each hunk, and commit only the changes you choose.

Suggestions never receive direct write access to your document. The review layer is the mutation boundary.

Your keys and model route stay part of the design

Cloud copies of documents and persisted analysis artifacts are encrypted on your device before upload. Local documents remain in browser storage and rely on your device and browser security. AI features need plaintext context while they run, and that context goes only through the route you selected.