Agents always launch inside a project context, opening in that project's directory and giving them access to your code and files.
How to launch
There are several ways to launch an agent:
- When enabled, use the New agent/terminal shortcut (
Cmd+Ton macOS andCtrl+Ton other platforms by default) to open the new-tab picker for the current project. - Right-click a project or process and choose Add > Agent.
- Use Agent with flags when you want to edit the full launch command before starting.
You can remap or disable the New agent/terminal shortcut under Settings > Hotkeys.
In the picker, project-runnable agent tools appear as items like New Claude agent. The tool and its matching installation must both be enabled, the installation health must be Ready or Not checked, and the project environment and root must be compatible.
Which project gets the agent
The agent launches in the currently selected project. If you want an agent in a different project, click on that project in the sidebar first, then, when enabled, use the New agent/terminal shortcut (Cmd+T on macOS and Ctrl+T on other platforms by default).
When you launch from a process context menu, Solo inserts the new agent next to that process in the same project.
Launch flags
The normal launch command is the tool command plus its default arguments from Settings -> Agents. Agent with flags opens a modal with that full command prefilled so you can change flags for just this launch. If you reuse the same flags regularly, save them as a named custom agent instead.
Examples of flags you might add are model or permission flags supported by the underlying CLI. Solo stores the resolved command on the new agent process.
The agent appears in the sidebar
Your new agent shows up in the Agents section of the project:
- An animated indicator shows it starting, then working.
- The default label comes from the tool and its runtime state.
- Selecting it opens the interactive session in the main pane.
The terminal pane
The terminal pane shows the agent's interactive session. You type prompts directly into the terminal and the agent's output streams back in real time.
Agents are fully interactive — you can have a back-and-forth conversation, ask follow-up questions, or give entirely new tasks.
Running multiple agents
You can run multiple agents at the same time. In free or expired license states, Solo allows up to 20 total stored processes across all projects, including agents. Validated licensed states are not subject to that Solo process-count cap.
For example, you can run:
- Multiple instances of the same tool (e.g., two Claude Code sessions working on different parts of the codebase).
- Different tools running side by side.
- Agents spread across multiple projects.
Each agent runs independently in its own terminal instance. Switching between them in the sidebar shows each agent's own conversation history.
For a lead coordinating implementation lanes, follow the agent orchestration workflow. When you want several read-only opinions on one frozen artifact, use an independent review panel.
Nested agents
If an agent spawns another agent through Solo, Solo can nest the child under the parent in the Agents section. Nesting only happens when both rows belong to the same project, and only when the parent relationship is clear.
Parent rows can be collapsed. When collapsed, the parent can show a (+N subagents) count beside its summary.
Stopping, resuming, or closing an agent
See Stopping and resuming agents when you want to end the process but keep the conversation available. See Closing agents when you want to remove the agent from Solo entirely.