Welcome to the Fleet blog
This is the home for practical, hands-on writing about running self-hosted Claude Code agents, the kind that stay up 24/7, remember what they did yesterday, and quietly get work done while you sleep.
Fleet is open-source and self-hosted: you run it on your own Mac, bring your own Claude subscription, and own your data. The posts here are written from that same vantage point: real setups on real hardware, not toy demos.
What we’ll write about
The blog is organized around a few recurring themes:
- Always-on agents. Keeping a Claude Code agent alive around the clock, scheduling recurring work, and recovering cleanly from crashes.
- Building with MCP. Servers, tools, channels, and how agents talk to the outside world.
- Self-hosting on a Mac. Running on your own hardware with a bring-your-own-subscription model instead of per-token SaaS.
- Agent memory & state. Making an agent remember across sessions.
- The economics & architecture. Cost trade-offs, comparisons, and the decisions behind a durable agent setup.
Why self-hosted
Most “AI agent” products are hosted services that meter you per token and hold your data. Fleet takes the opposite position: the agent runs on a machine you control, on a flat-rate subscription you already pay for. For always-on workloads, that math tends to favor self-hosting, a topic we’ll dig into with real numbers.
# The shape of it: an agent that just keeps running.
# (Setup details live in the docs, link below.)
fleet status
# ✔ atlas running uptime 31d
# ✔ echo running uptime 12d
# ⏸ nova idle next run 07:00Every guide here uses Fleet’s actual stack as the worked example, and links to the open-source repo so you can read the code yourself.
More soon. In the meantime, the fastest way to understand Fleet is to explore the platform or read the setup guide.