/swarm Then to open the dashboard n To create one. Every agent is different zot Subprocesses with their own model loop, their own persistent session file, and their own chat in the dashboard – but they all run in the same working directory as the host, so they see and edit the same files as you do. You keep working in your main session; Sub-agents work in parallel.
Agents edit the same files you do. they use the same read / write / edit / bash The device acts as the main agent against the host’s working directory. There is no per-agent worktree or branch. If you need parallel editing on different checkouts, set it up yourself git worktree Outside area.
Press enter On any row to open that agent’s transcript – a chat overlay with an always-on inline composer at the bottom, auto-follow streaming, and a busy spinner showing the agent’s current activity (thinking, tool: edit_fileetc.). type and hit enter To send a follow-up; esc Returns to the dashboard.
Simple shapes from the command line so scripts and headless flows work the same as dashboards:
/swarm # open the dashboard
/swarm new # spawn an agent
/swarm new --model gpt-5 # pin the new agent to a model
/swarm logs # jump straight into one transcript
/swarm send # send a follow-up without the dashboard
/swarm resume # pick a stopped agent to bring back
/swarm kill # stop a running agent (its state stays)
/swarm remove # delete the agent's session and state Agents are limited to the session that spawned them and are visible only in that session’s dashboard. They persist under ZOT restart $ZOT_HOME/swarm/agents/; Press R One to return to the same session and inbox socket on a different line so that the conversation continues where it left off.
Per-agent everything (session files, event logs, inbox sockets, meta) lives under $ZOT_HOME/swarm/agents/. The actual code of the agent is edited directly in your repo; track them normally git status / git diff.
/session export Does not bundle sub-agents – a .zotsession There is only the main chat transcript, and the state of a Swarm agent (session file, Unix-socket inbox) is machine-local and cannot round-trip through one. To share what an agent said, manually copy it from the transcript view.
with /settings -> With auto-swarm turned on, when a request naturally splits into independent parallel tasks, the main agent can also automatically fork to sub-agents. Look Settings for details.
<a href