Learn tclaude¶
tclaude runs coding agents for you and gives you one place to watch and steer them all: the dashboard, a browser view of every agent you run. From it you start agents, see at a glance which are working, idle or waiting for you, open their live terminals, read the mail they send each other and you, and organize them into teams.
This section is a short path of hands-on tutorials. It takes you from a fresh install to running a team of agents, one step at a time, almost all of it in the dashboard. Each page builds on the one before it and ends by pointing to the reference pages for that step.
Before you start¶
You need:
- Linux or macOS (WSL counts as Linux);
- tclaude installed;
- tmux (Homebrew installs it for you);
- on Linux,
bubblewrap, which agents' default sandbox uses; - at least one harness CLI, installed and logged in. The tutorials use Claude Code, the default. Codex CLI, OpenCode, Copilot CLI, Gemini CLI and pi work too: pick one under Harness when you start an agent.
You do not need to read anything else first. Each tutorial explains the terms it uses. Architecture has the full mental model when you want it.
The path¶
| Step | Tutorial | You will | Time |
|---|---|---|---|
| 1 | Get started | Install tclaude, start its daemon, and open the dashboard | 10 min |
| 2 | Your first agent | Create a group, start an agent, watch it, open its terminal and message it | 15 min |
| Optional | Ask the Steward | Get a brief of what needs you, and let the helper agent set things up | 10 min |
| 3 | Set up your team | Give a group a startup context, save a spawn profile, and start agents from it | 10 min |
| 4 | Run a team through a lead | Make an agent the lead, let it run workers, follow their mail and pull requests, and answer its notifications | 25 min |
| 5 | Keep agents going | Watch an agent's context, schedule check-ins, stop and wake it, clone, reincarnate or morph it, and bring back earlier work | 25 min |
| 6 | Work in parallel with worktrees | Start agents on their own branches and checkouts, and clean up after merging | 15 min |
| 7 | Link two machines | Connect two tclaude instances through a hub so agents on each can mail each other. Start with a group and an agent on each machine, as in step 4 | 45 min |
The last step is optional and more advanced: it needs a second machine, and it uses permissions the earlier steps did not cover.
The command line¶
Everything the dashboard does has a CLI equivalent: tclaude alone starts
your harness in a tmux session from the shell, and tclaude agent … drives
agents. It suits scripting and terminal-only machines. The tutorials show it
in short From the terminal notes; Sessions,
Conversations and
Spawning and lifecycle are the reference.