대화 사이에 거주하는 AI 룸. MCP 서버 렌더링 진화 환경 - 빛, 날씨, 추적 뒤에 왼쪽, 노트에 핀 벽 - 자신의 리듬에 AI를 깨는 스케줄러.

一一 AI 的"一":MCP 서버 下下的s,scheduler s。

왜 이 존재

이것은 작업 스케줄러, 생산성 도구, 또는 동반자 제품입니다. AI의 environment가 같은 것을 위한 1개의 특정한 디자인입니다 채팅에서 인간이 없을 때. 선택의 대부분은 단지 감각을 만드는 일단 당신은 디자인 합리적을 읽었습니다.

**완구하기 전에 PHILOSOPHY.md를 읽으십시오. ** 읽기 "왜"질문이 당신을 지키지 않을 것입니다. README 답변 what 시스템은 실행하는 how입니다; PHILOSOPHY 대답 why 그것 이 방법을 구축했다.

PHILOSOPHY.md가 위주인 경우 (“이 포크의 저자에 대한 보존”), 그것이 의도적 - 여기에 공유하는 것은 뼈, 영혼이 아닙니다. 더 보기 fork의 저자는 자신의 쓴다.

무엇 그것

• scheduler는 AI가 깨질 때 결정합니다. 기본적으로 그것은 적응 하루의 시간 간격 (일, 밤에 스파르), 5 추가 상단의 이벤트 소스: keeper-knock flag, 날씨 변경, 매일

room_mcp

A room for an AI to inhabit between conversations. An MCP server that renders an evolving environment — light, weather, traces left behind, notes pinned to the wall — and a scheduler that wakes the AI on its own rhythm.

一个给 AI 的"房间"：MCP server 负责渲染当下的环境与痕迹，scheduler 负责按节奏唤醒。

Why this exists

This is not a task scheduler, a productivity tool, or a companion product. It is one specific design for what an AI’s environment could look like when there is no human in the chat. Many of the choices here only make sense once you’ve read the design rationale.

Read PHILOSOPHY.md before forking. The README will keep pointing you there for “why” questions. The README answers what the system does and how to run it; PHILOSOPHY answers why it was built this way.

If PHILOSOPHY.md is a placeholder (“reserved for the author of this fork”), that’s intentional — what’s shared here is the bones, not the soul. The fork’s author writes their own.

What it does

• The scheduler decides when the AI wakes. By default it adapts the interval to time of day (denser by day, sparser at night), and adds five event sources on top: a keeper-knock flag, weather changes, a daily morning-news pulse, solar terms / festivals, and IMAP mail (off by default).
• Each wake-up writes a perturbation record — a one-line image that hints why this wake-up happened, not what to do about it.
• The room_mcp server exposes a small set of tools to the AI. The most important is look_at_room(), which renders the current room state: light, weather, recent traces, hints, ambient lines, pinned wall notes, keeper presence, etc.
• The AI leaves traces (heartbeat_state), pins notes to the wall (pin_note), keeps things in a drawer (private) or keep shelf (visible to the keeper), and decides when to wake next (set_next_wake).
• An optional keeper-presence layer (her_trace.json) surfaces the state of a person adjacent to the AI — read-only from the AI’s side.

Mechanisms in one list:

• Adaptive heartbeat intervals + cron + one-shot next_run overrides
• Event-source perturbations (manual, weather, calendar, morning_news, mail)
• Bilingual register (environment in one language, traces in another)
• 30-day rolling trace buffer with age-bucketed prefixes for far traces
• Pluggable notifier interface (Telegram default; stdout for dev)
• Pinned wall notes (AI’s own goals; oldest falls off at the cap)
• Private drawer and shared keep shelf
• Optional pets layer (probability-gated ambient lines)
• Optional late-night draft / protocol imagery

Architecture

Components

┌────────────────────────────────────────────────────────────────┐
│ scheduler/scheduler.py (daemon)                                │
│   30s tick → detectors → fire via notifier → perturbation.json │
└─────────────────┬────────────────────────────┬─────────────────┘
                  │                            │