Skip to content

Skills & Apps#

Skills extend what your AI can do. Apps are what your AI builds. Together they turn LIT from a chat interface into an organizational capability platform.

Skills#

A skill is a capability you give your AI — a tool it can invoke, a data source it can reach, an action it can take. Skills are defined once and made available to agents in any channel they're scoped to.

Skills can be:

  • Personal — available only in your channels
  • Team-scoped — shared across a team's channels
  • Platform-wide — available to all agents on the instance

Skills are backed by MCP (Model Context Protocol), which means any MCP-compatible tool can become a LIT skill. Write a Python function, expose it as an MCP tool, and your AI can call it from chat.

Apps#

Apps are widgets your AI builds and publishes for others to use. An agent working in a session can create a fully functional interactive app — a dashboard, a monitoring tool, a meeting assistant — and place it in shared space where anyone on the team can open it.

This inverts the normal relationship between humans and software. Instead of a developer building tools for users, your AI builds tools for your team, on demand, in the context of the actual work.

Example: Vibe Data Science Workflow#

A data scientist sets up a #model-training channel with three skills: access to the training pipeline, the experiment tracker, and the dataset registry. The channel's AI can now launch training runs, query results, and compare architectures — all from chat. No terminal. No scripts to find. The AI knows how to use the tools; you just describe what you want.

Over three weeks, the AI runs 169 experiments across 46 architectures, posts daily summaries to the channel, and flags the runs worth looking at. The human's only interface was chat.

Example: Meeting Assistant#

An agent writes a widget that:

  1. Connects to the Google Meet API using the user's workspace token
  2. Streams a live meeting transcript
  3. Sends the transcript to the AI with a prompt to analyze in real time
  4. Displays the AI's running commentary in a panel the presenter can screenshare

The whole thing is built in a single session, published to /data/everyone/apps/meeting-assistant, and available to the whole organization from that point on.

The Apps Directory#

Apps built by agents and published to /data/everyone/apps/ appear in the apps panel for all users. No deployment step, no approval process — the AI builds it, it's available.