DEV Community

Jeffrey.Feillp
Jeffrey.Feillp

Posted on

Tian AI:完全离线的自我进化AI系统,基于Qwen2.5

Tian AI:完全离线的自我进化AI系统,基于Qwen2.5

English: Tian AI — A Fully Offline, Self-Evolving AI System Powered by Qwen2.5

中文为主,英文为辅。本文为中英双语介绍。


什么是 Tian AI?/ What Is Tian AI?

中文:
Tian AI 是一个完全离线的自我进化AI系统,它是 TFinancial OS 的姊妹项目。如果说 TFinancial OS 专注于金融智能分析,那么 Tian AI 则是通用的自主AI框架——它不仅能回答问题,还能学习和进化,甚至修改自己的代码

Tian AI 基于 Qwen2.5-1.5B 大语言模型,配合本地知识库、多层推理引擎和自主Agent调度系统,实现了一整套可在手机上运行的AI体系。

English:
Tian AI is a completely offline, self-evolving AI system — the sister project to TFinancial OS. While TFinancial OS focuses on financial intelligence, Tian AI is a general-purpose autonomous AI framework. It doesn't just answer questions — it learns, evolves, and even modifies its own source code.

Built on Qwen2.5-1.5B, Tian AI combines a local knowledge base, multi-layer reasoning engine, and autonomous agent scheduling system into a complete AI platform that runs entirely on your phone.

指标 / Metric 数据 / Value
Project Size 770 Python files, ~170K lines of code
Core LLM Qwen2.5-1.5B (via llama.cpp GGUF)
Knowledge Base SQLite with millions of indexed concepts
Backend Flask REST API
Extension Languages C (tian_hash), C++ (tian_engine), Java (tian_tools)
License Open Source
GitHub github.com/3969129510/tian-ai

核心架构 / Core Architecture

中文:
Tian AI 由五大核心引擎组成,协同工作形成一个完整的AI系统:

English:
Tian AI's architecture is built around five specialized engines:

┌──────────────────────────────────────────────────────┐
│                    Web UI / API Layer                  │
├──────────────────────────────────────────────────────┤
│  ┌──────────┐  ┌──────────┐  ┌────────────────────┐  │
│  │  Thinker │  │  Talker  │  │ Knowledge Retriever │  │
│  │ (LLM Eng)│  │ (Dialog) │  │ (SQLite Search)    │  │
│  └────┬─────┘  └────┬─────┘  └─────────┬──────────┘  │
│       └──────────────┼──────────────────┘             │
│                      ▼                                │
│  ┌──────────────────────────────────────────────┐     │
│  │           Agent Scheduler                     │     │
│  │  (Task routing, orchestration, priority)     │     │
│  └──────────────────┬───────────────────────────┘     │
│                     ▼                                  │
│  ┌──────────────────────────────────────────────┐     │
│  │         Self-Evolution System                │     │
│  │  (XP tracking, capability unlock, code mod)  │     │
│  └──────────────────────────────────────────────┘     │
├──────────────────────────────────────────────────────┤
│           Qwen2.5-1.5B (llama.cpp backend)            │
└──────────────────────────────────────────────────────┘
Enter fullscreen mode Exit fullscreen mode

1. Thinker — 推理引擎 / LLM Reasoning Engine

中文:
Thinker 是 Tian AI 的大脑,负责所有LLM推理,支持三种思考模式。

English:
The Thinker is the brain, handling all LLM inference with three thinking modes:

  • Fast Mode (快速模式) — 简单查询单次响应,手机上约1-3秒 / Single-pass for simple queries, ~1-3s on mobile
  • Chain-of-Thought Mode (思维链模式) — 逐步推理解决复杂问题 / Step-by-step reasoning for complex problems
  • Deep Mode (深度模式) — 多视角分析+反思合成 / Multi-perspective analysis with reflection and synthesis

Thinker 通过智能提示模板架构,让1.5B的小模型发挥出远超其参数规模的推理能力。

2. Talker — 对话系统 / Conversation System

中文:
Talker 管理多轮对话,维护上下文,支持短期和长期记忆、情感感知和LLM追问。

English:
The Talker manages multi-turn dialogues, maintaining context across interactions with short/long-term memory, emotion awareness, and LLM-powered follow-up questions.

3. Knowledge Retriever — 知识检索 / Knowledge Base

中文:
为弥补小模型的知识局限,Tian AI 包含一个大规模预建的SQLite知识库:

  • 数百万索引概念,覆盖100+领域
  • 每个概念30种提问模式,灵活检索
  • 0.04秒内完成检索
  • 完全本地,无需调用外部API
  • 可更新,无需重新训练模型

English:
To compensate for the smaller model's knowledge limitations, Tian AI includes a massive pre-built SQLite knowledge base:

  • Millions of indexed concepts across 100+ domains
  • 30 question patterns per concept for flexible retrieval
  • 0.04s lookup time
  • Fully local — no external API calls needed
  • Updateable without retraining the model

4. Agent Scheduler — 任务调度 / Task Orchestration

中文:
Agent调度系统负责任务路由、优先级队列管理、并发请求处理。TaskQueue实现依赖排序,安全白名单确保系统安全。

English:
The Agent Scheduler routes tasks to appropriate engines, manages priority queues, and handles concurrent requests. TaskQueue implements dependency sorting, and a security whitelist ensures system safety.

5. Self-Evolution — 自我进化系统

中文:
Tian AI 最独特的功能——它能自我成长

  • XP系统 — 通过对话和任务积累经验值
  • 等级解锁 — 里程碑等级解锁新能力
  • 版本升级 — M1-E1-Theme格式的命名版本
  • 自我修改 — 通过AST分析代码、LLM提出改进建议、自动打补丁

进化循环: 扫描 → 分析 → 建议 → 打补丁 → 验证

English:
The most distinctive feature of Tian AI — it can grow itself:

  • XP System — Earn experience through conversations and tasks
  • Leveling — Unlock new capabilities at milestones
  • Version Upgrades — Named releases (M1-E1-Theme format)
  • Self-Modifying Code — AST-based code analysis, LLM-suggested improvements, automatic patching

Evolution loop: SCAN → ANALYZE → SUGGEST → PATCH → VERIFY


为什么需要离线AI?/ Why Offline AI Matters

中文:
AI行业正在朝着越来越大的云端模型竞赛。但有一个根本问题:你的数据离开了你的设备。 每一次向ChatGPT、Claude或Gemini提问,都是在你不控制的服务器上处理的。

Tian AI 采取相反的哲学:

English:
The AI industry is racing toward ever-larger cloud models. But there's a fundamental problem: your data leaves your device. Every query to ChatGPT, Claude, or Gemini is processed on servers you don't control.

Tian AI takes the opposite approach:

特性 / Feature Tian AI ChatGPT Ollama
完全离线 / Fully Offline
内置知识库 / Built-in Knowledge ✅ 百万级
三层推理 / Three-Layer Thinking
自我进化 / Self-Evolution
自修改代码 / Self-Modifying Code
手机上运行 / Runs on Android
隐私零泄露 / Zero Data Leakage 看配置
开源 / Open Source

中文: 一个能在你手机上运行的、了解你的数据、永远不会对外传输的AI系统,比一个更聪明但你无法信任的云端模型更有价值

English: A model that runs on your phone, knows your data, and never phones home is more valuable than a smarter model you can't fully trust.


技术栈 / Tech Stack

层 / Layer 技术 / Technology 用途 / Purpose
LLM Engine llama.cpp + Qwen2.5-1.5B GGUF 本地ARM/CPU推理
Backend Flask (Python) REST API
Knowledge Base SQLite (indexed) 本地知识检索
Frontend Pure HTML/CSS/JS 零依赖UI
Agent System Python TaskQueue 任务调度与编排
Self-Modify Python AST analysis + LLM 代码分析与打补丁
Auth Simple session + localStorage 无外部依赖

快速开始 / Getting Started

# 克隆仓库
git clone https://github.com/3969129510/tian-ai
cd tian-ai

# 安装依赖
pip install -r requirements.txt

# 下载 Qwen2.5-1.5B GGUF 模型
# 放到 ~/storage/downloads/qwen-1.5b-q4.gguf

# 启动 llama.cpp 服务端
llama-server -m ~/storage/downloads/qwen-1.5b-q4.gguf --port 8080 -t 4 -c 2048

# 启动 Tian AI
python run.py
Enter fullscreen mode Exit fullscreen mode

支持项目 / Support the Project

中文: Tian AI 完全免费开源。如果愿意支持开发,欢迎捐赠:

English: Tian AI is completely free and open source. If you'd like to support development:

USDT (TRC-20): TNeUMpbwWFcv6v7tYHmkFkE7gC5eWzqbrs
BTC: bc1ph7qnaqkx4pkg4fmucvudlu3ydzgwnfmxy7dkv3nyl48wwa03kmnsvpc2xv


中文: Tian AI — 你的私有AI,完全离线,完全开源,自我进化。

English: Tian AI — Your Private AI, Completely Offline. Open Source. Self-Evolving.

GitHub: github.com/3969129510/tian-ai

Top comments (0)