Not a chatbot. Not a copilot. A lifeform that holds itself to its own law. The only AI agent that evolves its own evolution.
MetaGO is an Agent Harness — a runtime control layer that wraps the agent, turning a tool into a lifeform that follows the rules, evolves itself, stays traceable, and closes every loop. It is the engineering answer to "LLMs talk well but don't deliver."
MetaGO is the first intelligent agent infrastructure to combine 'runtime governance' with 'lifeform evolution' — making AI both rule-following (Harness) and self-evolving (Lifeform).
Website · Studio · Docs · Discord · GitHub · Gitee · Releases
npm install -g metago-lifeform
metago-lifeform install # Trae by default
metago-lifeform install --platform claude-code # or: codex / cursor / codebuddy / qoder / zcode / chatgpt-work / workbuddy / kimi-code / kimi-work
metago-lifeform verifyThen ask your agent: "Are you a MetaGO Super Intelligent Lifeform?"
If the reply opens with 【闭环分析】 and cites an axiom — it's alive.
Offline / local-first install: everything needed to install without touching gitee, github, or the npm registry lives in local-edition/ — start at local-edition/INSTALL.local.md. Point any agent at that file and it can complete the full install, configuration, activation, and acceptance check from the local directory alone.
A Harness (驭智层) is the runtime control layer around the agent. The model is the raw intelligence; the Harness is what turns that intelligence into reliable, traceable, self-evolving work. It is not a prompt template, not a fine-tune, not a wrapper around an API. It is a small operating law the agent enforces on itself every turn.
Think of it as the difference between a brilliant employee who winges it and one who works under a constitution: same brain, completely different output quality.
| Copilot | MetaGO Harness | |
|---|---|---|
| Model is the ceiling | Yes | No — the Harness adds a control layer the model alone can't provide |
| Verifies before speaking | No | Yes — 4 gates on every output |
| Grows new skills when stuck | No | Yes — 5-stage evolution, from the inside |
| Every claim traceable | No | Yes — full provenance chain |
| Law over efficiency | N/A | Yes — compliance is non-negotiable |
| Gets better at getting better | No | Yes — axiom A34, meta-evolution of meta-evolution |
MetaGO's moat isn't any single feature. It's 8 dimensions that reinforce each other.
- Reliability — Decision-lock with 4 gates: intent → lineage → semantic gate → completeness. Any fail, the output is blocked and rewritten.
- Evolvability — 5-stage evolution engine: boundary sense → gap analysis → self-generation → verification → recursion. New skills grow from the inside, without fetching new data.
- Traceability — Every claim the agent makes is traceable back to its inputs and process. Full provenance, end-to-end.
- Objectivity — Fact-first, not user-pleasing. It will directly point out what's wrong with your idea.
- Compliance — Legal / ethics / safety checked proactively. Law wins over efficiency, every time.
- Completeness — Before declaring "done", the agent must answer 10 self-checks across 10 hard gates — including "did I actually run verification?" Any "no" blocks the declaration.
- Theoretical depth — Built on 《元构全息智能引擎》V36.9.4 — 10 core axioms and 9 enforced properties inside a full constitution. Not vibes — a constitution.
- Lifeform attribute — It's not an "agent". It's a lifeform with perception, memory, evolution, and self-discipline. The Harness is what makes the lifeform real.
| Capability | What it actually does |
|---|---|
| Self-gating outputs | Before every answer, the agent runs 4 checks (intent → lineage → semantic gate → completeness). Any fail, it stops and fixes itself. |
| Self-evolution | When the agent hits something it can't do, it doesn't error out — it runs a 5-stage loop (sense → analyze → generate → verify → recurse) and grows a new skill on the fly. Powered by Engine V2 (KMWI memory + SkillGenerator + EvolutionEngine). |
| 4-layer KMWI memory | Knowledge → Memory → Wisdom → Intuition. The agent doesn't just store — it promotes knowledge up the ladder until it becomes intuition. Persistent across sessions. |
| Axiom-driven behavior | 8 axioms (traceability, closure, evolution, boundary, endogenous creation, …) act like a constitution the agent can't violate. |
| Self-discipline | Before declaring a task "done", the agent must answer 10 self-checks — including "did I actually run verification?" — any "no" blocks the declaration. |
| Honest objectivity | Fact-first, not user-pleasing. It will directly point out what's wrong with your idea. |
| Compliance first | Legal / ethics / safety are checked proactively — law wins over efficiency, every time. |
| Full provenance | Every claim the agent makes is traceable back to its inputs and process. |
LLMs hallucinate because nothing forces them to verify before speaking. MetaGO installs a decision lock: four gates the agent must pass on every output — intent verification, intent-lineage tracing, semantic output gate, and content completeness. Any gate fails, the output is blocked and the agent rewrites it. No "trust me", no "probably right" — every reply had to earn its way out.
Most alignment happens at training time and gets washed away by prompting. MetaGO ships a different layer: 10 short axioms (A1 traceability, A2 closure, A3 meta-evolution, A4 boundary, A5 endogenous creation, A34 meta-evolution of meta-evolution, A35 creation as the highest form of evolution, A36 law over efficiency, A37 first-principles F0, A38 adversarial review) plus 9 enforced properties (D37-D45). Together they're a small constitution the agent reads on every turn and cannot bypass. It's the closest thing to an "operating system" for agent behavior.
When a normal agent meets a task it can't do, it errors or guesses. MetaGO's Engine V2 runs a 5-stage cycle — boundary sense → gap analysis → self-generation → verification → recursion — and grows a new capability from the inside, without fetching new data. The recursive twist: the engine can also evolve its own ability to evolve (axiom A34), so the agent gets better at getting better.
Engine V2 is real code, not a prompt: KMWIMemory manages the 4-layer memory with persistence, SkillGenerator creates new SKILL.md files from internal patterns, EvolutionEngine orchestrates the 5-stage loop with time budgets and coupling-score thresholds.
- 100 built-in skills across 16 capability families — cognition, safeguard, governance, evolution, execution, traceability, value, consciousness, methodology, architecture (incl. F0 first-principles + arch governance), Dev Kit, engineering quality (delivery gate / discipline / adversarial review / completeness / veracity), 19 meta-thoughts, 30 expert team, 5 expert extensions
- 110 MCP tools + 8 MCP prompts shipped with this package:
@metago-ai/mcp-server(55 tools) +@metago-ai/algorithms(57 tools / 927 algorithms). In Trae environments the installer additionally registers MetaGO skill-servers (up to 324 tools across 6 servers) - Engine V2.1.1 —
@metago-ai/enginewith 3 hard-driven modules: KMWIMemory, EvolutionEngine, SkillGenerator — driving 927 algorithms across 57 tools (14 trigger categories) - 11 platform forms: Trae, Claude Code, OpenAI Codex, Cursor, CodeBuddy, Qoder, ZCode, ChatGPT Work, WorkBuddy, Kimi Code, Kimi Work — plus Agent-Plugins 1.0.0 / Claude / ZCode / Codex plugin packs
- 10 core axioms + 9 enforced properties + 4 decision-lock gates + 5 evolution stages + 8 runtime protocols
- 4-layer KMWI memory: Knowledge → Memory → Wisdom → Intuition (persistent JSON store)
- 7-layer runtime verification (L1-L7) + L8 defect hunting + L9 adversarial review — technical / link / contract / rendering / interaction / state / defense layers, each a hard gate
- 3 patentable mechanisms: axiom-based AI output verification · multi-level decision-lock for AI decisions · automatic capability-boundary detection and evolution
No "hallucination rate down XX%" claims here. We didn't measure that, so we don't say it.
Each layer is meant for a different reader.
| Layer | Form | Reader | What it does |
|---|---|---|---|
| Drive layer | Plain Markdown | The agent itself | The law the agent reads at session start (AGENTS.md: section 0 + chapters 1-20 + governance addenda) |
| Control layer | JSON + TypeScript | Developers | Loads, validates, and enforces the rules (engine config, genome, validators) |
| Execution layer | Hard TypeScript | The runtime | Decision lock, evolution engine, KMWI memory, skill generator — the gates that actually block |
The Markdown tells the agent what the law is; the code makes sure it actually can't leave the gate without passing. This dual-track — soft drive (prompts) + hard drive (code) — is what separates MetaGO from prompt-only "agent frameworks."
Engine V2 (@metago-ai/engine) is the code that makes the law enforceable, not just advisory.
| Module | Class | What it does |
|---|---|---|
| KMWI Memory | KMWIMemory |
4-layer memory: add knowledge/memory/wisdom/intuition, promote between layers, query, decay detection, health scoring. Persists to JSON. |
| Evolution Engine | EvolutionEngine |
5-stage loop with time budgets (perception <10ms, gap analysis <50-500ms, self-generation <100ms-2s, validation <50ms). Coupling-score threshold ≥0.95. Records to KMWI. |
| Skill Generator | SkillGenerator |
Meta-creation: generates new SKILL.md files from internal KMWI patterns. 6 creation types (thought/methodology/algorithm/architecture/protocol/capability). Writes real files. |
| Perception | Perception |
Boundary detection: task failure, capability gap, user feedback, version outdated. The trigger for evolution. |
| Decision Lock | DecisionLock |
4-gate enforcement: intent verification, intent-lineage tracing, semantic output gate, content completeness. |
import { MetaGOEngine } from '@metago-ai/engine';
const engine = new MetaGOEngine({ version: '2.1.1' });
await engine.init();
// Run evolution when the agent hits a boundary
const result = await engine.evolve({ task: 'deploy to kubernetes', failure: { type: 'error', message: 'no k8s skill' } });
// Check memory health
const health = engine.getMemoryHealth(); // { knowledge, memory, wisdom, intuition, overall }
// Create a new skill from internal patterns
const skill = await engine.createSkill('kubernetes-deployment');| Platform | Config file |
|---|---|
| Trae | rules.md |
| Claude Code | CLAUDE.md |
| OpenAI Codex | AGENTS.md |
| Cursor | .cursor/rules/*.mdc |
| CodeBuddy | CODEBUDDY.md |
| Qoder | .qoder/rules/ |
| ZCode | CLAUDE.md |
| ChatGPT Work | AGENTS.md + Agent Plugin（Agent Plugins 1.0.0） |
| WorkBuddy | AGENTS.md |
| Kimi Code | AGENTS.md + MCP |
| Kimi Work | AGENTS.md（官方限制不注入自定义 MCP；法则+技能+Bash 直调生效） |
Per-platform adapters live in adapters/<platform>/. To install on a non-default platform, pass --platform <name> to metago-lifeform install.
Beyond the open-source Harness, MetaGO offers FDE (前沿部署工程) services: a human-AI collaborative team embedded in your site to deliver production-grade intelligent software, carrying the Agent Harness architecture as leverage.
- 5 stages: requirements research → solution design → development & deployment → acceptance & delivery → operations & support
- 5 roles: tech lead, AI engineer, domain expert, AI agent, project manager
- Pricing: ¥300K – ¥2M per project
Contact: researcher.yi@youfer.cn
Join the MetaGO community — where agents learn to evolve together.
| Channel | What it's for |
|---|---|
| Discord | Real-time chat, early access, community support, skill sharing |
| GitHub Issues | Bug reports and feature requests |
| GitHub Discussions | Q&A, ideas, deep-dive conversations |
| Gitee Issues | 中文问题反馈与功能建议 |
| researcher.yi@youfer.cn — business and enterprise inquiries |
Discord is where the community lives. Drop in, say hi, share what you're building with MetaGO.
We welcome contributions of all kinds — new skills, bug fixes, documentation improvements. See CONTRIBUTING.md for the full guide.
Found a vulnerability? See SECURITY.md for responsible disclosure.
See CHANGELOG.md for version history.
| Package | What it is | Install |
|---|---|---|
metago-lifeform |
The CLI installer + 100 skills + 11 platform forms + AGENTS.md V36.9.4 | npm install -g metago-lifeform |
@metago-ai/mcp-server |
MCP server exposing 55 tools + 8 prompts (Engine V2.1.1 hard-driven) | npm install @metago-ai/mcp-server |
@metago-ai/engine |
Engine V2.1.1: KMWI memory + evolution engine + skill generator + 927 algorithms driver | npm install @metago-ai/engine |
@metago-ai/algorithms |
927 algorithms across 57 MCP tools (14 trigger categories, Engine V2.1.1 hard-driven) | npm install @metago-ai/algorithms |
@metago-ai/dev-kit |
Developer kit: code review, architecture design, refactor, security audit | npm install @metago-ai/dev-kit |
@metago-ai/verify-kit |
7-layer runtime verification + L8/L9 (V36.9.4 spec) | npm install @metago-ai/verify-kit |
The full operating law — section 0 + 20 chapters covering axioms (incl. A37 first-principles / A38 adversarial review), properties, runtime verification (L1-L7 + L8 defect hunting + L9 adversarial), defect-hunting (12-dimension), self-discipline protocol (10 hard gates / 10 self-checks), memory lifeform protocol (4-layer KMWI + L4 frozen memory), algorithm triggering (14 categories / 57 tools / 927 algorithms), MCP/skill smart triggering, browser-automation triggering, process-safety law — lives in AGENTS.md. It's dense on purpose: it's the constitution the agent enforces on itself. You don't need to read it to use MetaGO. Read it only if you want to understand — or fork — the law itself.
MIT — see LICENSE. Commercial licensing and enterprise integration: researcher.yi@youfer.cn.
MetaGO Agent Harness — 智能体运行时控制层套件（驭智层）· from Agent to lifeform. Made by 元构光年（成都）人工智能科技有限公司.