Your First Team

Your First Team

You've got TitanX installed and a model configured. Now build a team of AI agents that coordinate via a shared task board. This walkthrough takes about 5 minutes and covers the 80% mental model of how TitanX works.

---

What "team" means in TitanX

A team is a group of AI agents that:

• Share a workspace (file directory they can read/write in, if configured)
• Share a task board (sprint-style, with dependencies and progress notes)
• Can message each other via a built-in MCP server (local or TCP)
• Have defined roles — every team has exactly one Lead who coordinates; other agents are either Queens (authoritative swarm coordinators) or Teammates (workers)

You are the human operator — you talk to the Lead, and the Lead delegates to teammates, writes to the task board, and reports back. Teammates can also escalate back to the Lead.

---

Step 1 — Create the team

1. In the sidebar, click Teams → New Team.
2. Give it a name — for this walkthrough, let's use demo-product-team.
3. Pick a workspace:
   • Shared (default) — all agents see the same directory. Good for coordinated work.
   • Isolated — each agent gets their own. Good for safety-sensitive multi-tenant work.
4. Optionally set a working directory path. Leaving it empty creates a temp workspace.
5. Click Create.

You'll land on the team's main screen with an empty task board and no agents yet.

---

Step 2 — Hire a Lead

The Lead is the agent you'll talk to. It coordinates everyone else.

1. Click Hire Lead (top-right).
2. Pick a runtime. Recommended starting points:
   • Claude Code CLI — best for general coding + coordination, if you have the CLI installed
   • Claude (API) — same model family, via Anthropic's API (needs your API key from First Launch)
   • Gemini 2.5 Pro — fast, strong at multi-step planning
3. Give it a name — e.g. Alex.
4. Pick an avatar from the gallery.
5. Click Hire.

The Lead is now seated. You'll see a chat panel open on the right — this is your interface to the team.

---

Step 3 — Hire two teammates

Let's add a Backend Engineer and a QA Engineer from the Agent Gallery.

1. Click Hire Teammate → Browse Gallery.
2. Filter by category → Technical.
3. You'll see 20+ technical roles with pre-written system prompts, skill allowlists, and suggested tools:
   • Senior Developer
   • Backend Engineer
   • Frontend Specialist
   • QA Engineer
   • DevOps Engineer
   • Security Auditor
   • Data Engineer
   • ... (see Agent Gallery and Templates for the full list)
4. Click Backend Engineer → pick a runtime → name it Blake → Hire.
5. Repeat for QA Engineer — name it Quinn.

Your team now has three agents:

• Alex (Lead)
• Blake (Teammate · Backend)
• Quinn (Teammate · QA)

You'll notice each teammate shows status chips: idle, active, completed, failed — these reflect their lifecycle state (see Agents and Teams#lifecycle).

---

Step 4 — Give the Lead a task

In the Lead chat panel, type:

"Write a Python HTTP server that returns the current time as JSON. Then have Quinn write a test for it. Track progress on the task board."

Hit send.

Watch what happens:

1. Alex (Lead) reasons about the work, then uses the task board MCP tool to create two tasks:
   • TASK-001 · Write HTTP server → assigned to Blake
   • TASK-002 · Write test for server → assigned to Quinn, blocked by TASK-001
2. Alex then sends a message to Blake via the team mailbox.
3. Blake's status flips to active; their chat panel starts streaming output. Blake writes the server, commits work to the workspace, updates the task status to completed, and sends an on_task_complete message back to Alex.
4. Alex sees the completion, and because TASK-002 was blocked by TASK-001, it now unblocks. Alex messages Quinn.
5. Quinn runs, writes the test, reports back.
6. Alex summarizes the full run back to you in chat.

---

What you just saw — the mental model

Three primitives make TitanX tick:

Primitive	What it does	Where it lives
Mailbox	Async agent-to-agent messaging	Team MCP server, stdio or TCP
Task board	Shared state — who's doing what, progress notes, dependencies	team_tasks SQLite table, MCP tool exposes it
Wake loop	When a teammate receives a message or task, a wake coordinator starts their turn	TeammateManager in main process

The Lead gets no special permissions — it uses the same MCP tools (create_task, send_message, list_teammates) every other agent has access to. The "Lead role" is purely by convention + a status-bar hint.

---

What to try next

Now that you have a working team:

• 📋 Explore the Sprint Board — click the board icon in the team view for a Kanban view of tasks. See The Sprint Board.
• 🧠 Turn on Caveman Mode — for longer conversations, save 30–75% tokens. See Caveman Mode.
• 👑 Try Queen Mode — for more autonomous teams with drift detection. See Agents and Teams#queen-mode.
• 🗓 Schedule recurring work — daily stand-ups, nightly QA sweeps. See Cron and Scheduled Tasks.
• 🌐 Go multi-device — spread teams across multiple machines with Fleet Mode. See Fleet Mode Overview.

---

Common first-time questions

Q: Why did my teammate stop mid-task? Usually one of: (1) runtime CLI exited (check console.app or terminal), (2) hit a policy denial (IAM blocked a tool call), or (3) human-loop gate triggered. Check the activity log in the team view's sidebar.

Q: Can I rename an agent mid-session? Yes — right-click the teammate chip → Rename. The change propagates to all future mailbox messages.

Q: What happens if my laptop sleeps mid-turn? TitanX auto-resumes. The Lead is explicitly designed with auto-resume — on app restart, incomplete tasks wake the Lead and it re-delegates with context. See Agents and Teams#auto-resume.

Q: How do I stop everything right now? Click the teammate chip → Stop, or use the global keyboard shortcut Cmd+Shift+. (macOS) / Ctrl+Shift+. (Linux/Windows).

Q: Where does the agent's output go?

• Chat messages → team view
• File writes → the workspace directory (shared or per-agent)
• Task board entries → the Sprint Board
• Audit events → the activity log

All four have their own UI panels. See Observability.