Moonweave AI

Long-lived intelligence, made legible.

We engineer AI systems that preserve meaning, state, authority, and evidence across long horizons.
_{构建在长期运行中仍能维持语义、状态、权责与证据一致性的 AI 系统。}

Public Atlas ↗  ·  Repositories ↗  ·  Research Archive ↗  ·  Governance ↗

Agent Systems   Semantic Infrastructure   Embodied Intelligence   Governed AI

---

A model can look capable in a demo. A system must remain coherent across sessions, tools, teams, and years.
模型可以在一次演示中显得聪明；系统必须在跨会话、跨工具、跨团队、跨年份后仍保持同一性。

Moonweave AI is a research-oriented organization building long-lifecycle AI architecture: agents and semantics form the cognitive layer, infrastructure provides the operating substrate, and controlled embodiment connects intelligence to virtual and physical environments.

Our projects are designed as one connected system rather than a pile of unrelated repositories. Shared semantics, explicit coordination, persistent state, replayable evidence, and governance are treated as engineering primitives, not documentation added after the interesting part has already escaped into production.

Project constellation · 项目体系

Cantilune 0.x · agent orchestration A general-purpose language and control substrate for visible, evolvable coordination across agents, tools, people, services, permissions, sessions, and scarce resources. Repository · 简体中文	Moonweave Agent Ontology ontology workspace · explorer A governed, single-source model of agent systems with recursive YAML authority, deterministic JSON projections, verification, and a force-directed graph explorer. Repository · Explorer
Ontotect public preview · npm An evidence-driven ontology engineering skill suite for coding agents: route, build, review, repair, refactor, validate, govern, and release RDF/OWL/SHACL systems. Repository · npm · 简体中文	Moonweave AI Governance active · organizational operating system Principles, roles, RFC and ADR workflows, engineering gates, quality evidence, security boundaries, collaboration protocols, and knowledge practices. Repository · 中文 · 日本語
Axiolune · 同枢 pre-alpha · financial ontology platform An ontology-centered financial research and operations platform connecting market data, research, models, portfolios, risk, decisions, and controlled execution through one governed semantic model. Repository	VIREA research preview · embodied motion A verifiable retargeting system that turns heterogeneous human-motion sources into one auditable VRM/glTF humanoid contract with replayable artifacts and portable playback. Repository · 中文文档
Awesome Agentic AI Blog Archive living archive · research A curated OpenAI and Anthropic corpus reorganized into thematic reports for studying agent architecture, memory, tools, evaluation, safety, infrastructure, and deployment. Repository · 简体中文	Kaguya Moonweave Project public atlas · EN / 中文 / 日本語 The public research portal for Moonweave AI: project definition, architecture, roadmap, research notes, Moonlog, media provenance, safety, and governance. Website · Source

Engineering doctrine · 工程原则

Across repositories, Moonweave AI follows four practical constraints:

• Define meaning once. Shared domain contracts should outlive individual models, frameworks, languages, and storage engines.
• Make coordination inspectable. Ownership, authority, resources, sessions, state transitions, and topology changes should remain explicit.
• Publish evidence, not atmosphere. Status, limitations, provenance, validation, and unresolved risk belong beside every capability claim.
• Treat embodiment as separately authorized. Physical action requires bounded permissions, simulation, human oversight, reversibility, and fail-safe controls.

Start here · 从这里开始

Explore the system Read the manifesto, architecture atlas, five-year roadmap, research notes, and public governance. Open the Public Atlas ↗	Use an engineering tool Inspect the 20 focused Ontotect entries, then install the suite into a supported coding-agent workspace. npx @moonweave-ai/ontotect list
Study modern agent systems Use the research archive and its thematic reports as a structured reading map. Browse the archive ↗	Collaborate with context For cross-project proposals, begin with the governance model, quality gates, and RFC process. Read Governance ↗

Contributing · 参与协作

Moonweave AI welcomes focused, evidence-backed contributions in research, agent systems, ontology engineering, evaluation, infrastructure, documentation, and embodied intelligence.

Start with the target repository's README and contribution rules. Cross-project proposals should identify the problem, scope, evidence, affected contracts, risks, acceptance criteria, and expected artifacts. Large or irreversible changes should follow the governance repository's RFC and review process rather than arriving as an heroic 4,000-line surprise.

Public boundaries · 公开边界

Moonweave AI builds independent, original systems. Public assets must be original, licensed, or clearly referenced; provenance and synthetic-media status should remain visible. Embodied capabilities are developed under explicit safety boundaries and human supervision. Licenses and maturity levels are repository-specific, so review each project's README, LICENSE, and third-party notices before reuse.

Original systems. Shared meaning. Visible evidence. Controlled action.
_{原创系统 · 统一语义 · 可见证据 · 受控行动}

English · 中文 · 日本語

_{© Moonweave AI · Built in public, governed for the long term.}