
molis-ai/molis-work
Developer toolsこのページは AI により自動翻訳されています。技術的な詳細は原文をご確認ください。 原文を見る
Goal truth source shared by people and Agent Runtimes
概要
Molis Work is a local-first work platform: a plugin base with many plugins, where you and the AI Runtimes you use work on the same goals, sessions, feeds, code, documents and calendar, each kept by the plugin that owns it. The platform is one resident host per Home, a shared action catalog with grants, the plugin runtime, the workbench shell, and one AI runtime. Every project has Goals and can add Sessions, Inbox, Feed, Schedule, Results and the Coding family as needed; personal plugins such as Jelly (calendar and notes), Todo, Pages, Lingguang and Cognia work in every project. Each plugin registers its capabilities once in the action catalog, where pages, the Assistant, workflows and external MCP clients call them as granted. The Goals plugin is for a boring way long-running work fails: a new Session cannot see the last one, the original outcome drifts as local decisions pile up, and “done” is a sentence with nothing to check. What is missing is not a smarter model.
README
以下は原文 README の抜粋です。原文を見る。
Molis Work
English | 简体中文
Molis Work is a local-first work platform: a plugin base with many plugins, where you and the AI Runtimes you use work on the same goals, sessions, feeds, code, documents and calendar, each kept by the plugin that owns it.
The platform is one resident host per Home, a shared action catalog with grants, the plugin runtime, the workbench shell, and one AI runtime. Every project has Goals and can add Sessions, Inbox, Feed, Schedule, Results and the Coding family as needed; personal plugins such as Jelly (calendar and notes), Todo, Pages, Lingguang and Cognia work in every project. Each plugin registers its capabilities once in the action catalog, where pages, the Assistant, workflows and external MCP clients call them as granted.
The Goals plugin is for a boring way long-running work fails: a new Session cannot see the last one, the original outcome drifts as local decisions pile up, and “done” is a sentence with nothing to check. What is missing is not a smarter model. It is one project record every Runtime can read: the accepted Goal, how it was split, what is blocked, and what counts as done.
Molis Work keeps that record locally. Codex, Claude Code, OpenCode, or another connected Harness updates the same Goal. You confirm material changes. You can see how far the work has got without asking the model to recap.
It does not ship a model: you configure models and keys in Settings, and the Assistant and plugins such as Coding, Schedule and the plugin studio run through the Home’s one Prologue Runtime. External Harnesses still read and write through MCP; Molis Work does not dispatch work for you.
The longer derivation (WeChat article, link forthcoming): [placeholder — 公众号文章待发布]. Draft: From conversational facts to ledger facts.
Three ways to use it
Same project, three surfaces.
Desktop
A native macOS window: focus one Goal, then open a terminal that stays bound to it. Molis Work also has a clickable status icon in the macOS menu bar. Click it for the current project, focused Goal, status, and next action.
推奨ツール
別のキーワードを試すか、フィルタを外してください。
インストール
npx skillfish add molis-ai/molis-work