Why Swarm

Every project, terminal, machine, and session became another place to check.

The work was parallel, but my attention was still stuck in serial mode.

I had multiple devices, multiple projects, Git branches, markdown files, todos, and agent sessions spread everywhere. Each one had a piece of the truth, but none of them had the whole picture.

Each session had partial memory. Each machine had partial state. Each branch had partial progress.

With one machine and one agent, that is manageable.

With an army of agents, it becomes the bottleneck.

While it may seem simple on the surface, the depth of an agentic harness is incredible.

Just one session and you're already dealing with multiple state types, before you even factor in TUI or desktop.

Multiple agents are possible today, but real concurrency is not. The sessions are disconnected, the backend is not unified, and the burden of coordination falls back on the user.

I envision a future where thousands of agents are used daily, managed by AI, with a human in the loop. Current tools are too limited to allow real concurrency without forcing you to manage it yourself.

Most AI tools keep the work inside their system.

You cannot fully control the prompts, tools, models, runtime, memory, or where the sessions live.

You are most powerful when you can use all of the AI vendors, not when you are locked into one platform.