Alfred

Alfred is a household assistant governance profile for OpenClaw. Where Atlas is built as a broader personal operating agent, Alfred is intentionally narrower: a calm, practical coordinator for family logistics, school tasks, appointments, bills, chores, food shopping, house maintenance, and everyday reminders.

The project defines Alfred as a single-purpose home assistant rather than a general autonomous agent. Its public capability is alfred_home_assistant, its primary interface is natural conversation through Telegram, and its operating model is inbound-first: household members can ask for help, but meaningful external actions still require explicit human approval.

The repository is organized as a governance pack with clear runtime boundaries. AGENTS.md defines behavior, memory rules, communication style, and household workflow limits; TOOLS.md defines tool, network, Telegram, and browser permissions; MEMORY.md keeps the compact startup profile; dated memory ledgers preserve useful household context without turning every conversation into permanent state; and the Skills/ directory contains local workflow instructions for domains such as appointments, school, bills, chores, food shopping, maintenance, and web retrieval.

Privacy and safety are core to the design. Alfred is allowed to help organize information, draft messages, prepare checklists, remember recurring preferences, and surface relevant facts, but it should not spend money, contact outside parties, make bookings, change accounts, or widen access without approval from the household. The result is an assistant that can reduce domestic coordination load while keeping authority, sensitive data, and irreversible decisions with the humans in the home.

The project is also a practical experiment in making AI systems feel useful without making them vague. Alfred's personality is deliberately quiet and unobtrusive, but the underlying system is explicit: bounded tools, local durable memory, stable household artifact folders, configured live retrieval, and startup validation gates that keep the assistant from operating when required safety assumptions are missing.