Every action of the Vopit window has a vopit command, with machine-readable output. A coding agent with shell access can therefore open a project, shape the team, choose connections, plan the Board, create Cranks, hold meetings and run requests. It goes through the same engine and the same checks as you: your license, the project trust and every permission.
Set it up once
- Install Vopit and its command: in PowerShell on Windows, irm https://vopit.fr/install.ps1 | iex; on Linux, curl -fsSL https://vopit.fr/install.sh | sh. If Vopit is already installed on Windows from vopit.fr, use Settings → Terminal → Add to terminals instead.
- Sign in once yourself with vopit login. The agent then uses this session; it never needs your password.
- Open the project yourself and review it before trusting it: vopit open <folder>, then vopit open <folder> --trust.
- Choose the connection the team uses, for example vopit connection use claude-cli --model opus, and check it with vopit connection test claude-cli.
Brief the agent
vopit guide prints a manual written for models: how Vopit works, the conventions every command follows, typical workflows and every command with its options. Point the agent to it, and write your own rules next to it, in the instructions file the agent reads (such as CLAUDE.md or AGENTS.md).
This project uses Vopit, a team of AI agents, through the `vopit` command.
- Run `vopit guide` once to learn the commands. Always add `--json`.
- Send work with `vopit -p "<task>" --json` and read the result.
- Exit code 3 means an agent waits for a permission: tell me what it asks.
Do not run `vopit approve` yourself.
- Never use `--yes` (deleting, replacing the team, discarding Git changes) without asking me.The request loop
vopit -p "Add input validation to the signup form" --json
# exit 3: the JSON gives permission.id and requestId
vopit approve <permission-id> once # you, or the agent if you allow it
vopit wait <request-id> --json # again until exit 0 (done) or 4 (partial)
vopit show <request-id> --json # result, changed files, steps| The agent can | With |
|---|---|
| Shape the team | agent create · agent generate · agent set · team generate · team import |
| Configure connections | connections · connection use · connection set · connection key (from stdin) |
| Plan and run work | board generate · board add · board start · board auto · request · wait |
| Automate | crank flow · crank create · crank enable · crank run |
| Discuss | meeting start · meeting say · meeting capture · meeting apply |
| Inspect | status · requests · show · memory · git · folders |