What you actually need before your first agent session

An orchestrator for coding agents is a tool for running things you already have. That is the whole idea, and it is also the thing that catches people out on day one: you install it, you open it, the list on the left is empty, and the next step is not obvious because it happens somewhere else.

So here is the short answer to what you need, before you need it.

One: a coding agent, installed on that machine

DevThrottle is vendor-neutral. It runs the coding agents you already use, each in its own session on its own repository, and you keep your own subscriptions and keys - it does not resell you an agent and it does not sit between one and its service.

Eight agents have a built-in driver: Claude Code, Pi, Codex, Gemini, OpenCode, Cursor, Grok and GitHub Copilot. There is no waiting list and no gate, but there is one step that is yours: you install that agent's own command-line tool yourself. DevThrottle detects it, it appears as a row in settings, and from there it is one of the options when you start a session.

An agent whose tool is not installed on the machine is simply not in the list. That is true of every one of them. If the New Session window says there are no agents configured, this is the step that is missing - not something that went wrong with the install.

Two: a repository worth pointing it at

A session lives its whole life in one repository, running the one agent you picked. So the choice matters slightly more than it looks: you can rename a session whenever you like, but a different repository or a different agent means a different session.

The repository list fills itself in two layers - the ones you have opened before, most recent first, and below them everything found under the root folders you gave the setup wizard. If you pointed the wizard at the folder where your code lives, the list is already full on your first visit. Only when there is genuinely nothing does it offer a browse button.

Three: a decision about approval prompts

This is the one that decides whether a fleet actually runs. An agent that stops to ask permission on every tool call is fine when you are watching it and useless when you are running six. So a freshly detected agent is set up to work without stopping for approval wherever its own command line offers a way to say so.

Six of the eight do offer one. Two do not, because their tool has no such option to give: one only trusts files inside the project, and one keeps permissions in its own configuration file rather than on the command line. That is a property of those tools, not a setting hidden somewhere.

There is a checkbox in the New Session window for this, and one rule worth knowing: the agent's own settings entry decides first. If that entry has already settled the permission question, either way, the checkbox changes nothing. It only matters for an entry that settles nothing, and there it is the whole decision.

What happens if you skip one

Nothing half-happens, which is the point. Start Session only comes alive once an agent and a repository are both chosen, so a grey button means one of the two is missing. And if the agent's command-line tool is not actually installed on that machine, you are told before anything launches rather than left looking at a session that died quietly.

Ten minutes, most of them the agent's rather than yours. The walkthrough with every screen in it: start your first session. The full list of agents and what each one launches with: supported agents.