Skip to content
LVIS AI
EN|KO

Architecture · Diagrams

Architecture visuals

This page turns the system's big picture into pictures. Click any card to zoom in for a closer look.

click to zoom
stack · flow · permissions · lifecycle
Zooming in
Clicking a diagram card zooms it into a lightbox that fills nearly the whole screen. The text scales up along with it.

Diagram 01

The system at a glance — four layers

App · plugins · local storage · server. Calls and storage flow from top to bottom.

Diagram 02

The flow of a single message

Starting from user input through to the result returning to chat. Risky tools go through user confirmation along the way.

Diagram 03

Permission branching — risk level × tool type

Branches into auto-run, a confirmation card, or a dialog depending on the tool's risk level (low/medium/high) and type (read/write/execute/network).

Diagram 04

Plugin lifecycle — now and next

The top row is the current state, the bottom row is stages to be added later. The dotted line shows where they hook in.

Future vision diagrams

The three diagrams below translate items from the roadmap into pictures. They show designs yet to be added, not current behavior.

Diagram 05 · vision

Capability Pack — publish once, install once

Bundles plugins · agents · MCP · Skills into a single package for publishing, so users can install everything in one step.

Diagram 06 · vision

Sub-agent delegation — autonomous execution after consent

For complex requests, once the user allows delegation, a sub-agent autonomously calls tools. All results are collected back into chat.

Diagram 07 · vision

Federation — delegating work to another user

Hands off work to a person on another host and receives the response. Trust is expressed through key exchange.
Current vs. future distinction
Diagrams 01-04 show the current state; Diagrams 05-07 show designs yet to be added. Parts yet to be added may include names that don't yet exist in the code, so please don't cite them as fact.