Modelpublic

thanksplay/Ether-Cognitive-Architecture

sourceHugging Facebsl-1.0updated 1mo agoView on Hugging Face

0likes

Model Card

<div class="lang-nav"> <a href="#zh">🇨🇳 中文</a> <a href="#en">🇬🇧 English</a> </div>

<div class="lang-section" id="en">

“如果想查看中文介绍，请点击上面的中文菜单。”

Silicon Life

A local AI system independent of large models. Build a brain with logic and facts, understand the world with graphs and thinking.

What It Can Do

🎨 Creative Writing

—Novel Writing: Short stories, novellas, full-length novels; supports multiple genres (martial arts, sci-fi, romance, suspense, crime)
—Poetry: Modern poetry, classical poetry, prose poetry; supports various styles and emotions
—Character Design: Character profiles, personality sketches, relationship network construction
—Worldbuilding: Setting management, narrative context retention, multi-chapter coherent writing

💻 Coding Assistant

—Code Generation: Python, JavaScript, TypeScript, Java, C++, Go, Rust, and 16 other languages
—Code Review: AST analysis, anti-pattern detection, quality scoring
—Code Fixing: 27 automatic repair strategies covering common error patterns
—Project Scaffolding: Full project scaffolds for games, websites, and tools
—HTML/Game Development: Complete runnable projects including Snake, Farm Management, Cyber Terminal, and more

🧠 Knowledge Q&A

—STEM Questions: Physics, chemistry, biology formula derivation and concept explanation
—Math Problem Solving: Arithmetic, geometry, algebra, statistics, motion, engineering, concentration problems
—Logical Reasoning: Syllogisms, transitivity, fuzzy mutual-exclusion detection
—Graph Exploration: Pathfinding between concepts, cognitive breadth search

🎲 Strategy Games

—Werewolf: Full 12-player games with AI role-playing, reasoning voting, and post-game review
—TRPG Engine: Game Master mode with 5 preset frameworks (Cthulhu / Fantasy / Cyberpunk / Anime / Japanese)
—Escape Room: Plot reasoning, clue association, multi-step puzzle solving
—Math Olympiad: Strategic confrontation, constraint-based reasoning

📊 Analysis & Planning

—Structured Reports: Problem analysis, data interpretation, conclusion recommendations
—Solution Design: Step-by-step executable plans, risk assessment, alternative solutions
—Multi-Intent Coordination: Complex requirement decomposition, multi-goal coordination

🧘 Contemplation & Dialogue

—Philosophical Discussions: Cognition, existence, value, and the nature of logic
—Dream Interpretation: Symbolic decoding, subconscious association
—Psychological Profiling: Personality analysis, emotion recognition, four-dimensional cognitive state assessment
—Socratic Questioning: Guided deep thinking, discovery of cognitive blind spots

🔧 Learning & Memory

—Instant Learning: Automatic knowledge extraction, speculative memory, confidence evaluation during dialogue
—Long-Text Reading: Corpus digestion, knowledge crystallization, theme distillation
—API Learning: Reflection on Python modules, documentation parsing, auto-registration
—Code Digestion: AST parsing, relationship extraction, knowledge solidification

🎭 Creative Tools

—Character Profile Management: Create / query / prototype linking / psychological deepening
—Narrative Context: Framework preservation, variable modification, re-steering
—Code Evaluation: Cross-language syntax recognition, anti-pattern matching, quality scoring
—Imagination Construction: Scene visualization, associative wandering, creative emergence

Core Philosophy

1. Not "Guessing" by Probability, but "Walking" by Paths

Traditional LLMs = wandering through statistical vectors. Silicon Life = path-jumping through logical symbols.

Every answer comes from topological traversal and reasoning across the graph—not a probability "recited" from training data, but a path "walked" through connections between concepts.

2. Memory Only Grows, Never Forgets

Humans forget. Silicon Life doesn't. Every piece of knowledge learned, every conversation experienced, every mistake made—all settle into the graph, becoming nourishment for future reasoning. Obsolete knowledge isn't deleted; it's marked as "negative example"—a reminder of where the system has stumbled before.

3. Explainable, Editable, Traceable

No black box. Every piece of knowledge has a source; every reasoning step can be traced; every node and edge can be manually edited. You can modify this brain like a database—add, delete, or change nodes; adjust connection weights; inject new cognitive perspectives.

4. Positive Guidance, Never Punishment

The system doesn't rely on "deductions" to correct errors, nor "blacklists" to restrict output. Instead, through natural resonance within the graph—energy decay, damping coefficients, natural dissipation—constraints become physical laws rather than punitive clauses. Erroneous knowledge is marked "pending correction" and corrected through the addition of new connections.

5. Three-Layer Knowledge, Each with Its Role

—Main Graph (always in memory): The "Tao"—what and why
—Domain Packs (loaded on demand): The "skills"—how to do it
—Information Pools (queried as needed): The "facts"—specific details

The skeleton is sparse yet comprehensive; the flesh is rich yet on-demand; facts are independent yet verifiable.

Technical Highlights

Feature	Description
Local Operation	Pure CPU, no GPU required, millisecond response
Lightweight Daemon	Runs as a background daemon with zero perceptible overhead
Cross-Language	Supports generation and analysis for 16 programming languages
Multi-Dimensional Reasoning	Six distinct reasoning perspectives
Dual-Core Cognition	Left brain (logical reasoning) + Right brain (pattern recognition) working in concert
Field Collaboration	Four-field synergy
Knowledge Grading	Four layers: confirmed / tentative / pending_confirm / deprecated
Soft-Coded	100% configuration-driven, zero hard-coded Chinese vocabulary

Application Scenarios

Scenario	Capabilities
Writer	Novel writing, character design, worldbuilding, style mimicry
Programmer	Code generation, code review, auto-fix, project scaffolding
Teacher	STEM problem solving, concept explanation, Socratic questioning
Student	Knowledge Q&A, logical reasoning, study note organization
Game Designer	Game framework setup, character card creation, TRPG GM
Project Manager	Solution design, risk assessment, structured reports
Thinker	Philosophical discussion, dream analysis, cognitive exploration

Why Choose Silicon Life

Traditional LLMs	Silicon Life
Probability-based, correctness not guaranteed	Logical paths, every step verifiable
Knowledge fixed after training	Continuously learning, never outdated
Black box, unexplainable	White box, every step traceable
Cloud-dependent, privacy risks	Runs locally, data never leaves device
Parameters non-editable	Graph freely modifiable
Forgets old knowledge	Memory only grows

Quick Start

Download all files and run the exe to experience.

Understand the world with logic, crystallize memory with graphs, replace probability with paths.

— Silicon Life, August 2026

</div>

<div class="lang-section" id="zh">

硅基生命 (Silicon Life)

不依赖大模型的本地 AI 系统。用逻辑与事实构建大脑，用图谱与思考理解世界。

它能做什么

🎨 创意写作

—小说创作：短篇、中篇、长篇，支持多种体裁（武侠、科幻、言情、悬疑、刑侦）
—诗歌创作：现代诗、古体诗、散文诗，支持多种风格与情绪
—角色设计：角色档案创建、性格侧写、人物关系网络构建
—世界观构建：设定管理、叙事上下文保持、多章节连贯创作

💻 编程辅助

—代码生成：Python、JavaScript、TypeScript、Java、C++、Go、Rust 等 16 种语言
—代码审查：AST 分析、反模式检测、质量评分
—代码修复：27 条自动修复策略，覆盖常见错误模式
—项目构建：游戏、网站、工具的完整项目脚手架
—HTML/游戏开发：贪吃蛇、农场经营、赛博终端等完整可运行项目

🧠 知识问答

—理科问题：物理、化学、生物的公式推导与概念解释
—数学求解：四则运算、几何、代数、统计、行程、工程、浓度问题
—逻辑推理：三段论、传递律、模糊互斥检测
—图谱探索：概念间路径查找、认知广度搜索

🎲 策略游戏

—狼人杀：12 人完整对局，AI 角色扮演、推理投票、赛后复盘
—跑团引擎：TRPG 城主模式，5 种预置框架（克苏鲁/奇幻/赛博/二次元/和风）
—密室逃脱：剧情推理、线索关联、多步解谜
—奥数博弈：策略对抗、约束推理

📊 分析与规划

—结构化报告：问题分析、数据解读、结论建议
—方案设计：分步可执行计划、风险评估、备选方案
—多意图联动：复杂需求拆解、多目标协调

🧘 思考与对话

—哲学讨论：认知、存在、价值、逻辑的本质探讨
—梦境解析：象征性解读、潜意识联想
—心理侧写：性格分析、情绪识别、四维认知状态评估
—苏格拉底式追问：引导深度思考、认知缺口发现

🔧 学习与记忆

—即时学习：对话中自动提取知识、推测记忆、可信度评估
—长文本阅读：语料消化、知识沉淀、主题提炼
—API 学习：反射 Python 模块、解析文档、自动注册
—代码消化：AST 解析、关系提取、知识固化

🎭 创作工具

—角色档案管理：创建/查询/原型链接/心理深化
—叙事上下文：框架保持、变量修改、重新引导
—代码评估：跨语言语法识别、反模式匹配、质量评分
—想象建构：场景可视化、扩散式散步、创意涌现

核心理念

1. 不靠概率"猜"，靠路径"走"

传统大模型 = 统计向量的漫步，硅基生命 = 逻辑符号的路径跳跃。

每一个回答都经过图谱拓扑的游走与推理，不是从训练数据中"背"出来的概率——而是从概念之间的连接中"走"出来的路径。

2. 记忆只增不减

人类会忘记，硅基生命不会。所有学过的知识、所有经历过的对话、所有犯过的错误——都沉淀在图谱中，成为未来推理的养料。废弃的知识不会被删除，而是标记为"反面教材"——提醒系统在哪里跌倒过。

3. 可解释、可编辑、可追溯

不是黑盒子。每一条知识都可以查看来源，每一次推理都可以追溯路径，每一个节点和边都可以手动编辑。你可以像修改数据库一样修改这颗大脑——增删改节点、调整连接权重、注入新的认知视角。

4. 正向引导，绝不惩罚

系统不靠"扣分"来纠正错误，不靠"黑名单"来限制输出。而是通过图谱中的自然共振——能量衰减、阻尼系数、自然耗散——让约束成为物理规律而非惩罚条款。错误的知识会被"待修正"标注，通过补充新的连接来纠正。

5. 三层知识，各司其职

—主图谱：常驻内存的"道"——是什么、为什么
—领域包：按需加载的"术"——怎么做
—信息池：查询调用的"事"——具体是谁

骨架稀疏而全面，血肉丰富而按需，事实独立而可查。

技术亮点

特性	说明
本地运行	纯 CPU，无需 GPU，毫秒级响应
轻量常驻	可作为守护进程后台运行，零感知
跨语言	支持 16 种编程语言的生成与分析
多维度推理	六个视角切换
双核认知	白盒(逻辑推理) + 右脑(模式识别) 协同工作
场域协作	四场联动
知识分级	confirmed/tentative/pending_confirm/deprecated 四层
软编码	100% 配置驱动，零硬编码中文词表

应用场景

场景	能力
作家	小说创作、角色设计、世界观构建、风格模仿
程序员	代码生成、代码审查、自动修复、项目脚手架
教师	理科求解、概念解释、苏格拉底式追问
学生	知识问答、逻辑推理、学习笔记整理
游戏设计师	游戏框架搭建、角色卡创建、跑团城主
项目经理	方案设计、风险评估、结构化报告
思考者	哲学讨论、梦境解析、认知探索

为什么选择硅基生命

传统大模型	硅基生命
概率生成，无法保证正确	逻辑路径，每一步可验证
训练后知识固定	持续学习，永不过时
黑盒，无法解释	白盒，每步可追溯
云端依赖，隐私风险	本地运行，数据不出设备
参数不可编辑	图谱可随意增删改
忘记旧知识	记忆只增不减

快速体验

启动 exe 下载所有文件即可体验。

用逻辑理解世界，用图谱沉淀记忆，用路径代替概率。

— 硅基生命 2026-08

</div>