Overview

The AI can use tools while answering you. It reads your screen, fetches links, and works with your connected apps.

When you ask a question, the AI can decide to use a tool before answering: read the window you're working in, fetch a link you mentioned, check your clipboard, or search your past Kerlig chats. It can chain several tools in one answer, and you see each step live as it happens.

When tools are active

Tools run only when you ask by hand, using the Ask action or any follow-up question you type in an ongoing chat. Quick actions like Fix spelling or Translate never use tools, so they stay as fast and cheap as before. You can enable tools for any action (or disable them for Ask) with the Use Tools switch in action preferences.

What the AI can use

Two kinds of tools are available:

• Built-in tools are read-only tools that ship with Kerlig: reading app windows, fetching web pages, the clipboard, and your local chat history.
• MCP servers bring tools from apps and services you connect yourself using the Model Context Protocol.

Every tool is governed by approvals and permissions. Nothing runs without a policy you control.

Model support

Tools require an AI model that supports tool calling. Most current models from OpenAI, Anthropic, Google, and other major providers do. When the selected model can't use tools, Kerlig shows "Tools off for this model" and answers normally without them.