AgentFleet – Prompt-driven fleet ops

Patch many small servers by prompting an agent instead of SSH-ing into each one

Problem

Individuals and small teams running a dozen non-critical servers waste time patching and editing config host by host. Some already prompt Claude or Codex to operate the fleet and save hours, but lack a safe connector and worry an agent will delete files or misconfigure something.

Solution

A lightweight connector that, only on explicit approval, lets a local AI agent reach a declared host inventory: task templates for audits and patches, highlighted commands before execution, post-run logs, rollback scripts, plus directory and command boundaries.

Users

Indie developers and small ops teams managing a few to a few dozen VPS, self-hosters

MVP

Host inventory and key management; audit/patch task templates; pre-run diff preview; command allowlist and directory sandbox; execution log with one-click rollback

Revenue

Open-source core; hosted subscription priced per host (~$9/mo for 5 hosts) with task audit, team sharing, and scheduled audits

Why now

Prompt-based ops is already validated in practice, but safety boundaries and rollback are unsolved; the tool runs locally without needing data centers or licenses