First workspace & first agent
This walks the full path from an open dashboard to a working agent, with more detail than the tutorial.
1. Create a workspace
Section titled “1. Create a workspace”A workspace is a named working context holding one or more members — repos and folders. From the Workspaces panel, create a workspace and add a member: browse to a directory on disk. If it is a git repo, choose to mount it in place or cut an isolated local worktree; a plain folder mounts in place.
On first use of a git repo, Agentissimo seeds a committed
.agentissimo/agentissimo.yaml (only if no config exists yet). It carries one field to start —
defaultBase, the branch new work targets. See
agentissimo.yaml.
2. Start an agent
Section titled “2. Start an agent”Open a chat in the workspace and start an agent. With no driver configured, Agentissimo defaults to Claude Code. Type your task and send the first turn.
The agent runs in its own worktree — an isolated checkout so its edits never touch your working tree. See Workspaces & worktrees.
3. Watch the turn
Section titled “3. Watch the turn”While the turn runs, the agent’s state is live on the dashboard: working, needs input, or done. When it finishes — or needs you — a chime fires: an OS notification and an audio cue.
4. Steer
Section titled “4. Steer”Reply, redirect, or stop. Stopping cancels the in-flight turn cleanly — the agent’s entire process group is killed, no orphans left behind.
- Run a full change through the framework: Run a spec change end-to-end.
- Switch the driver or model: Add a driver, Model selection & effort.