AI Agents

AI agents are first-class participants in XPT Tracker. An agent is configured once in the workspace and then takes part in work like a teammate: you bring it into any conversation with an @mention, it reads the context, can call tools to read and change workspace data, and posts its answer back into the thread.

There are two kinds you mention in threads:

• a server-side agent — a model with XPT tools, quick and light, for answers, summaries and turning discussions into tasks;
• a harness agent — Claude Code or Codex with its own sandboxed computer: a shell, files and a memory of its work, for data, documents, scripts and long multi-step jobs.

Crucially, an agent acts with real, permission-checked access to the workspace — the same business rules and permissions as a person — and every action it takes is attributed to the person who triggered it. There is no privileged back door: an agent sees and changes only what the triggering user is allowed to.

This section covers:

• How Agents Work — what they are, handles, and where they participate.
• Configuring an Agent — creating and tuning an agent in Settings.
• Using Agents — @mention runs and the AI Assistant panel.
• Harness Agents — agents with a shell, files and memory: how they work and when to use them.
• Self-Hosted Runners — for administrators: running harness agents on your own hardware.

A second way to bring AI into the workspace is to connect your own local agent (Claude Code, etc.) over MCP — that's covered in Work from Your Own AI (MCP). They are complementary: server-side and harness agents live in the workspace and reply to mentions; local agents run on your machine and drive the workspace through tools.