Skip to content

Xandria agents ​

An agent in Xandria is an AI that works inside your library. It searches your books, reads the passages it needs, cites the page behind each point, and can draft Notes, build Canvas boards and file books into Collections, with your approval.

There are two ways to work with one:

  • The built-in Agent, which you chat with in Xandria.
  • An outside agent you already use, such as Codex, which connects to your library through MCP.

Agents run in Xandria for Windows, macOS and Linux. They aren't part of the phone and tablet apps.

The built-in Agent ​

Xandria doesn't come with an AI service or an AI subscription of its own. The Agent runs on the model you choose:

Runs onGood forGuide
A local model with OllamaNothing leaves your computerSet up AI
An OpenAI-compatible providerA provider and key you already haveSet up AI
The Codex CLIYour existing Codex sign-in and plan (Windows, macOS, Linux)Use the Codex CLI
The OpenCode CLIYour existing OpenCode setup (tested on Linux)Use the OpenCode CLI

Then ask your library a question, or let the Agent edit Notes and Canvas.

Outside agents through MCP ​

If you already work with an agent such as Codex, turn on External control in Settings → AI → Permissions and add Xandria to it. The agent can then search your books, read passages and draft notes, with the same approvals as the built-in Agent. See Connect outside agents.

You stay in control ​

  • Edits to Notes, Canvas and Collections wait for your Approve unless you turn on automatic approval yourself. Anything applied can be undone.
  • An agent sees only what you can see.
  • Conversations stay on your computer.

More in How AI works in Xandria.

Common questions ​

Does Xandria work with Codex? Yes, in two ways: the built-in Agent can run on the Codex CLI, and Codex can use your library as an outside agent through MCP.

Does Xandria support MCP? Yes. Xandria is an MCP server for outside agents while it's open and you're signed in.

Do I need to pay for Xandria's AI? No. You bring the model: a local one, a provider you already pay for, or your Codex or OpenCode sign-in.