Getting Started

Introduction

xiantong brings AI agent coding and computer-use capabilities to your entire digital workflow. It improves both coding and non-coding workflows through an enhanced interface, review flow, and automated execution. Built on the same foundation as Claude Code, it extends model reasoning across your tools: GitHub, Linear, Slack, xiantong documents, local files, and any API you can connect.

What Makes xiantong Different#

Familiar interface#

The inbox works like a combination of email and task manager. Each conversation has a status, persists between sessions, and can be organized however you want. Review agent plans before execution in a document-like interface.

Connect to anything#

Sources give the agent access to your tools — MCP servers, REST APIs, or local files:

Source Type	Examples	What You Can Do
MCP servers	GitHub, Linear, xiantong	Integrate with issue tracking, docs, and repos
REST APIs	Any service with an API	Query and mutate data from internal tools
Local files	Obsidian vaults, code repos	Work with files on your machine

Multitask naturally#

Start multiple tasks, jump between conversations with full history preserved, and run long tasks in the background.

Explore, then execute#

Start in Explore mode for read-only research and planning. Review proposals before changes happen, then switch to Execute when ready.

Sessions become documentation#

Every conversation captures decisions, trade-offs, and implementation details. Share sessions with your team so others understand what changed and why.

Work from anywhere#

Run a remote server and access your workspaces from the desktop app, browser, or CLI.

Fully customizable#

Themes, skills, statuses, and permissions are backed by configuration files that the agent can help edit.

xiantong vs Claude Code#

Aspect	Claude Code	xiantong
Interface	Terminal CLI	Desktop app
Data sources	MCP servers	MCP + REST APIs + local files
Session organization	Continue with `-c`	Inbox with custom workflow statuses
Config scope	Per-project `.claude/`	Multi-workspace isolation
Permissions	Fixed mode behaviors	Customizable rules per workspace/source
Remote access	SSH + terminal	Remote server with desktop, browser, and CLI clients

Key Concepts#

Sources connect xiantong to external data. Workspaces isolate sources, skills, and sessions. Permission modes control what the agent can do.

Mode	Behavior
Explore	Read-only research and analysis
Ask to Edit	Prompts before changes
Execute	Full autonomous execution

Sessions are persistent conversations organized like an inbox. Skills are reusable instructions for specialized behaviors.