Documentation · Feature guide · Workspace

Agents and sessions

Status: In the app

The director agent plans and delegates; specialist agents build. Agent definitions, memory and reports live in files; sessions are started and watched from inside the app.

Contents

What it does

Team screen: architect, director, macOS / Swift engineer and QA agent cards
Team: the project’s agents and their workload (interface in Turkish).
  • The Team screen lists the project’s agents: role, open tasks, reports and skills; base agents come from the library.
  • The Director Island is the capsule at the top of the window: it shows how many Claude Code sessions and agents are running; click it for the Chat and Terminal panel and the sub-agent tree.
  • The AI Session window shows duration, tokens and estimated cost for every turn; when the budget runs out the session stops and you grant more.
  • The Library holds ready agent and skill definitions; a lesson learned in one project is promoted to the library.

Step by step

  1. Open Team in the project menu; open an agent card for its definition, memory and reports.
  2. To hand a task to an agent, press Start Session in the task detail; pick the AI tool (Claude Code, Codex) and the model.
  3. Watch running sessions from the island at the top; click it to open the Director panel.
  4. Follow tokens and cost per turn in the session window; change the permission mode if needed.
Director Island panel: session list, sub-agent tree and Chat / Terminal tabs
Director Island open: sessions and the sub-agent tree.
AI Session window: two turns with duration, tokens and cost, permission mode picker
AI Session: tokens and cost per turn (recorded sample session).

Related hx commands

# a project-specific agent from a base agent
hx new agent perf-engineer --base macos-engineer

# copies agents, skills, rules and hooks from the library into the project
hx sync

# an agent picks up its first pending instruction
hx inbox take --agent designer

Frequently asked questions

Is model usage included in the subscription?

No. Agents run on your own Claude Code and/or Codex account; model usage is billed there.

What if a session stops halfway?

The task, acceptance criteria and agent memory are files; a new session continues where the last one stopped.

Where do I set the budget?

You set the session budget when starting it; when it is spent the session stops and an extra-budget request comes to you.

Need help?

If you did not find your answer here, write to us and we will reply.

info@hexabitlabs.com Try the demo first