Skip to content
LVIS AI
EN|KO

Roadmap

Static Integration Today, Autonomous Collaboration Next

LVIS's next direction is growing plugins from simple tool-calling modules into companions capable of their own workspace, autonomous action, and delegation between people. This is a declaration of direction, not a schedule commitment.

v1 → v4 evolution
Declaration — not a schedule commitment

Vision

Every change falls under one of the six directions below. A single version can advance several directions at once, and each direction matures gradually across multiple versions.

Plugin to Connector + UI

Grows beyond simple tool-calling into a first-class connector that also provides its own workspace and live widgets.

Plugin = Sub-agent

Plugins become small agents that call multiple tools in a row on their own judgment. The user sets the scope of delegation.

Making use of idle time

Quietly handles indexing, summarizing, and prep work while the user isn't looking, then yields immediately once they return.

Capability Pack

Combines plugins, agents, MCP, and skills — currently published separately — into one bundle for single-step publish and install.

Richer automation

Expands beyond simple time/end triggers into conditional automation that combines email, meeting, and calendar events.

Hooks — standard points to plug into

Exposes standard points in the host flow — tool execution, permission grants, routine firing, and more — where plugins can safely step in.

Flow by version

Starting from v1 "Foundation" through to v4 "Frontier." Each version carries a single tone.

  1. v1

    Foundation — Elevating plugins to connector status

    • Plugin Workspace — richer UI

      In progress

      Until now, a plugin could only have one small area inside the sidebar. The next step expands this so plugins can have their own workspace, multiple panels, and draggable widgets.

    • Capability Pack — publish as a single bundle

      Planned

      Today, plugins, agents, MCP, and skills are each published and installed separately. Once a publisher bundles them into one package, users can install everything in a single step.

      View related diagram

    • Richer automation triggers

      Exploring

      Right now there are only two triggers: "on end" and "scheduled time." This expands into conditional automation combining email, meeting, and calendar events.

    • Expanding plugin lifecycle stages

      Planned

      Lets plugins define their own behavior at more points in time — when installed, when activated, when a token expires, and more.

      View related diagram
  2. v2

    Autonomous — plugins become small agents

    • Plugin = Sub-agent (autonomous execution)

      Exploring

      Plugins currently only respond when called. In the next step, once the user consents, a plugin becomes a small agent that calls multiple tools in sequence on its own judgment.

      View related diagram

    • Making use of idle time

      Exploring

      While the user isn't looking at LVIS, it quietly handles work like indexing, summarizing, and preparing ahead of time, then yields immediately once the user returns.

    • Live widgets

      Planned

      Beyond static cards, widgets that update in real time can appear inside the chat body — for example, a meeting transcript widget showing STT results live.

    • Standard hook points into the host flow

      Exploring

      Exposes standard points in the host flow — right before/after tool execution, when permission is granted, and more — where plugins can safely hook in.

      View related diagram
  3. v3

    Open — connecting to the outside world

    • Local LLM fallback path

      Exploring

      Can automatically switch to an on-device LLM (such as Ollama) so core flows keep working even when the network is down.

    • Federation — connecting to other hosts

      Exploring

      Currently this only works within a single organization. This expands it so work can be delegated to, or messages exchanged with, other LVIS users.

      View related diagram

    • Automatic classification of external MCP servers

      Planned

      Automatically assesses the risk of tools from external MCP servers so operators can approve them quickly.

  4. v4

    Frontier — safer, and closer at hand

    • Stronger plugin isolation

      Exploring

      Introduces a sandbox that caps memory, CPU, and network usage so third-party publisher packages can run more safely.

    • User automation performance board

      Planned

      See on one screen how many of the suggestions you received you accepted, and which automations saved the most time.

    • Mobile and external-tool companions

      Exploring

      Provides a mobile app and IDE/launcher extensions for quickly handling just the board, inbox, and approval responses.

How to read this page
The v1–v4 entries on this page are a direction of evolution, not a promise. As priorities shift, an item's status (exploring / planned / in progress / shipping) is updated freely. Please don't cite this as the basis for an external contract or SLA.
Previous

Permission Model