agent-fleet grew out of wanting to work with many agent sessions at once. My CLI sessions quickly became difficult to keep track of, so I built a shared overview. The system was originally made for Claude Code.
After I was later banned from Claude Code, I added support for Codex and OpenCode. That let me start sessions with different agents and bring them together in the same interface.
Each agent runs in its own terminal session, which I can reattach to from the CLI or work with through the GUI. When a client has a question, it is sent to the server over MCP and shown as a dialog in the GUI. That makes it easier to follow several sessions without constantly switching between terminals.
The project is older now, but the need behind it comes up again in several of my newer tools: working with multiple agents while keeping track of what they are doing and where they need my input.