B站 AI 学习互动机器人 — AI 自动刷视频、学知识、评论互动、私信回复、自我进化
版本: 3.1.2 | License: MIT
此插件改写经过原作者xiaoyaya bilibili_learning_bot同意
| 功能 | 说明 |
|---|---|
| 📺 智能视频浏览 | AI 驱动 B站推荐流浏览，自动判断内容价值（评分 / 收藏 / 投币 / 点赞） |
| 📚 知识库系统 | 自动归档高质量视频，3 层分类 + 语义检索 + 复习回顾 |
| 💬 评论互动 | 真实/模拟评论模式，AI 深度回复，支持图片分析 |
| 📩 私信处理 | 自动回复粉丝私信，持久上下文 + 长期记忆，支持节奏控制 |
| 📡 实时监听 | 独立监听引擎，只盯私信 + 评论实时 AI 回复，不刷视频不耗精力 |
| 🔔 @通知响应 | 视频下评论 "@bot 总结这个视频"，自动识别并总结回复 |
| 🧬 日记与自我进化 | 行为日志 + AI 自我反思 + 人格动态进化 |
| 🎙️ ASR 语音识别 | 视频语音转文字（FunASR / Whisper，可选安装） |
| 🤖 Agent 技能系统 | 自主规划目标 → 搜索 B站 → 看视频 → 总结知识，全自动闭环 |
| 🎓 知识辅导 | AI 讲解 / 问答 / 二次创作 / 生成 HTML 学习卡片 |
| 🎨 视频→网页 | 视频生成 PPT 风格 HTML，19 种视觉风格，支持 Claude 主题 |
| 📊 思维导图 & Word 导出 | 视频一键导出 .mindmap.html 与 .docx 文档 |
| 🔍 深度研习 | 长视频多章节深研，证据链式总结（services/deep_dive.py） |
| 🎯 智能兴趣引擎 | 多维度评分 + 同义词 + 排除词 + 灵光一闪探索 + PsychoProfile 同步 |
| 😊 AI 心情系统 | 动态心情影响互动风格，支持自定义 |
| 🏆 干货点赞回顾 | 定期回顾收藏的干货视频，AI 复习（services/like_review.py） |
| 🔔 本地提醒 | 桌面通知 + 待办提醒（services/reminders.py） |
| 🛡️ 安全审查 | 关键词过滤 + 政治敏感拦截 + 提示词注入防护 + 操作风控 |
| 🔄 备用 API 降级 | 主 API 连续失败自动切换备用提供商 / 备用模型 |
| 维度 | v3.0.2 | v3.1.2+（当前 3.1.2） |
|---|---|---|
| 代码规模 | 77 个 Python 文件 / ~34k 行 | 113 个 Python 文件 / ~54k 行（+47%） |
| 数据目录 | 项目内 Data/（打包/升级易丢） |
✅ %LOCALAPPDATA%\BiliLearn（打包产物零隐私数据，升级不丢） |
| 人格管理 | 简单 prompt 配置 | ✅ Web 可视化多人格（创建 / 编辑 / 激活 / 删除），key 与显示名双匹配 |
| HTML 渲染 | 各模块各自维护模板 | ✅ services/html_renderer.py 统一渲染（阅读页 / 幻灯片 / 导出） |
| 服务模块 | 12 个 | ✅ 32 个（新增深度研习、测验生成、思维导图、Word 导出、本地收藏、点赞回顾、提醒、RAG 问答、平台适配、代理配置、版本历史…） |
| 评论回复 | 基础回复 | ✅ 顶层/子回复路由修复、12006 失效处理、AI 选择失效 ID 跳过 |
| 监听引擎 | 基础轮询 | ✅ 上下文合并、超时跳过、-509 退避、网页日志可视化 |
| 开放平台桥接 | ❌ | ✅ ob_bridge/（开放平台鉴权、AB 测试、审计） |
| 备份与恢复 | 手动导出 | ✅ 分组备份（设置 / 记忆 / 知识 / 产物）+ 恢复 |
| 测试 | 43 个 pytest | ✅ 181 个 pytest（319 passed 全量发布验证） |
| 稳定性修复 | — | 人格持久化、Cookie 校验、风控、多实例锁、AI 降级冷却、上下文截断保护 |
详细演进见 CHANGELOG.md。
├── main.py # 🚀 主入口（CLI 交互菜单 + 自动化启动）
├── BiliLearn.spec # 📦 PyInstaller 打包配置
├── api/ # 🔌 B站 API 层（客户端 / 登录 / 字幕 / 节流）
├── brain/ # 🧠 核心大脑（Mixin 组合：主循环 / 视频理解 / AI 调用 / 会话）
├── cli/ # 💻 命令行菜单
├── core/ # ⚙️ 配置 / 全局变量 / 用户数据路径 / 恢复出厂
├── knowledge/ # 📚 知识库（分类 / 搜索 / 浏览 / 复习 / 自定义）
├── persona/ # 🎭 人格 + 心理画像引擎
├── security/ # 🛡️ 内容安全审查
├── services/ # 🔧 32 个服务（深研 / 测验 / 思维导图 / Word / 兴趣引擎 / RAG…）
├── ob_bridge/ # 🌉 开放平台桥接（鉴权 / AB 测试 / 审计）
├── xingye_bot/ # 🤖 扩展组件（LLM / 状态 / 记忆 / 进化 / ASR / 网格帧）
本项目仅供学习与个人研究使用。请遵守 B站用户协议与相关法律法规，合理控制互动频率，任何使用后果由使用者自行承担。
MIT © xiaoyaya191
#英文版
Bilibili AI Learning Bot — An AI that auto-watches videos, learns knowledge, interacts via comments, replies to DMs, evolves itself, with a built-in Web admin panel, and one-click Windows EXE packaging.
Version: 3.1.2 | License: MIT
| Feature | Description |
|---|---|
| 📺 Smart Video Browsing | AI-driven browsing of Bilibili's recommendation feed, automatically judging content value (rating / favoriting / coin / like) |
| 📚 Knowledge Base System | Auto-archives high-quality videos, with 3-tier categorization + semantic search + review & recap |
| 💬 Comment Interaction | Real/simulated comment modes, AI in-depth replies, with image analysis support |
| 📩 Direct Message Handling | Auto-replies to fan DMs, with persistent context + long-term memory, and pacing control |
| 📡 Real-time Listening | Standalone listening engine that only watches DMs + comments for real-time AI replies — no video watching, low resource use |
| 🔔 @ Mention Response | Comments like "@bot summarize this video" are auto-detected and summarized/replied |
| 🧬 Diary & Self-Evolution | Behavior logs + AI self-reflection + dynamic persona evolution |
| 🎙️ ASR Speech Recognition | Video speech-to-text (FunASR / Whisper, optional install) |
| 🤖 Agent Skill System | Autonomously plans goals → searches Bilibili → watches videos → summarizes knowledge, a fully automated loop |
| 🎓 Knowledge Tutoring | AI explanations / Q&A / secondary creation / generates HTML study cards |
| 🎨 Video → Webpage | Generates PPT-style HTML from videos, 19 visual styles, with Claude theme support |
| 📊 Mind Map & Word Export | One-click export of videos to .mindmap.html and .docx documents |
| 🔍 Deep Study | Multi-chapter deep study of long videos, with evidence-chain summaries (services/deep_dive.py) |
| 🎯 Smart Interest Engine | Multi-dimensional scoring + synonyms + exclusion words + "eureka" exploration + PsychoProfile sync |
| 😊 AI Mood System | Dynamic mood affects interaction style, customizable |
| 🏆 High-Value Like Review | Periodically reviews favorited high-value videos, AI recap (services/like_review.py) |
| 🔔 Local Reminders | Desktop notifications + to-do reminders (services/reminders.py) |
| 🛡️ Safety Review | Keyword filtering + political-sensitivity blocking + prompt-injection protection + operation risk control |
| 🔄 Fallback API Degradation | Auto-switches to backup provider / backup model after consecutive primary API failures |
| Dimension | v3.0.2 | v3.1.2+ (current 3.1.2) |
|---|---|---|
| Code Size | 77 Python files / ~34k lines | 113 Python files / ~54k lines (+47%) |
| Data Directory | In-project Data/ (lost on packaging/upgrade) |
✅ %LOCALAPPDATA%\BiliLearn (packaged artifact contains zero private data, survives upgrades) |
| ✅ Dashboard / Bot Control / Real-time Listening / Persona Management / Knowledge Tutoring / Deep Study / Backup Restore | ||
| Persona Management | Simple prompt config | ✅ Web visual multi-persona (create / edit / activate / delete), key + display-name dual matching |
| HTML Rendering | Each module maintained its own templates | ✅ services/html_renderer.py unified rendering (reading page / slides / export) |
| Service Modules | 12 | ✅ 32 (added deep study, quiz generation, mind map, Word export, local favorites, like review, reminders, RAG Q&A, platform adaptation, proxy config, version history…) |
| Comment Replies | Basic reply | ✅ Top-level/child reply routing fix, 12006 failure handling, AI skips invalid selected IDs |
| Listening Engine | Basic polling | ✅ Context merging, timeout skip, -509 backoff, web log visualization |
| Open Platform Bridge | ❌ | ✅ ob_bridge/ (open-platform auth, A/B testing, audit) |
| Backup & Restore | Manual export | ✅ Grouped backup (settings / memory / knowledge / artifacts) + restore |
| Tests | 43 pytest | ✅ 181 pytest (319 passed full release verification) |
| Stability Fixes | — | Persona persistence, Cookie validation, risk control, multi-instance lock, AI degradation cooldown, context truncation protection |
See CHANGELOG.md for detailed evolution.
├── main.py # 🚀 Main entry (CLI interactive menu + automated startup)
├── api/ # 🔌 Bilibili API layer (client / login / subtitles / throttling)
├── brain/ # 🧠 Core brain (Mixin composition: main loop / video understanding / AI calls / session)
├── cli/ # 💻 Command-line menu
├── core/ # ⚙️ Config / globals / user data path / factory reset
├── knowledge/ # 📚 Knowledge base (categorization / search / browsing / review / custom)
├── persona/ # 🎭 Persona + psychological profile engine
├── security/ # 🛡️ Content safety review
├── services/ # 🔧 32 services (deep study / quiz / mind map / Word / interest engine / RAG…)
├── ob_bridge/ # 🌉 Open-platform bridge (auth / A/B testing / audit)
├── xingye_bot/ # 🤖 Extension components (LLM / state / memory / evolution / ASR / grid frames)
└── dev_refs/ # 📖 Secondary-development reference docs
Q: Where is the data stored?
Source version: project root Data/; Web/EXE version: %LOCALAPPDATA%\BiliLearn (Cookies, API Keys, knowledge base, and QR codes are all local only; the packaged artifact contains no private data).
Q: The bot exits immediately after startup, log shows ImportError?
Check whether you're using a clean Python environment. If PYTHONPATH points to another Python's site-packages (e.g., multiple Pythons installed), import PIL may load a mismatched Pillow. Run echo %PYTHONPATH% before running — empty is safest.
Q: AI call reports 'ascii' codec can't encode...?
Check whether config.json's api.vision_api_key / unified_api_key was written as a placeholder like "[hidden]" (don't write back a desensitized config export). Clearing that field falls back to unified_api_key.
Q: Persona save says "does not exist"?
Caused by old-version data where the persona storage key didn't match the display name. 3.1.2+ already supports key/display-name dual matching; if it still happens, restart the panel to load new code, or delete Data/web_personas.json to re-migrate from personas.json.
Q: Why doesn't the exported config include Cookie and API Key?
Export has two modes: desensitized export (default, API Key / Cookie replaced with [hidden], safe to share with others) and full export (includes real Key and login Cookie, for your own migration backup only, filename carries _full suffix). The web panel asks which to choose on export; the CLI menu enters f for full export. A full export imported to a new machine has login state and AI config working immediately.
Q: After importing someone else's config backup, AI is all broken, reports 'ascii' codec can't encode?
On backup export, API Key / Cookie is desensitized to [hidden]; old versions would import and overwrite real config with the placeholder. 3.1.2 stable has fixed this: on import it auto-filters [hidden] (keeps existing value if present, otherwise deletes the field and you must refill). Already-affected users please manually edit %LOCALAPPDATA%\BiliLearn\Data\config.json and replace the [hidden] in unified_api_key / vision_api_key with the real Key.
Q: Port already in use?
Default 18083; auto-increments if occupied. Or set WEB_PORT=xxxx && python web_panel.py.
The project author knows well that "the tool is innocent, but misuse is culpable." The following disclaimer stacks as many layers as needed — please read each item:
- Unofficial: This project has no relationship with the official bilibili (B站), is not an official release, and is not endorsed or responsible for by Bilibili. All trademarks and names belong to their respective owners.
- Personal learning & exchange only: This project is for personal learning purposes only, to study technologies such as HTTP / MCP / data processing. Any form of commercial use, profit-making, large-scale batch scraping, attacks, or abuse of Bilibili services is prohibited.
- Legal risk at your own expense: Bilibili's interfaces and terms of service may change at any time, and Bilibili has taken legal action against similar reverse-engineering projects (e.g., bilibili-api). This project is built on public interfaces and does not guarantee long-term availability; any disputes, account bans, or legal liabilities arising from using this project are borne solely by the user.
- Account security:
SESSDATAis the highest-privilege credential of a Bilibili account; this project stores it only in your local user directory. Never publicly share QR-login screenshots, auth.json, or Cookie contents — leaking them is handing your account to someone else. - Content copyright: Extracted subtitles, danmaku, comments, etc. are copyrighted by the original authors and Bilibili, for personal reading/learning only — do not repost, redistribute, or use commercially.
- Stability & availability: This project is provided "as is", without any express or implied warranty. Bilibili redesigns, risk control, network conditions, and other factors may all cause it to fail; when interfaces fail, follow the README to re-scan or fix it yourself — the author does not promise a fix time.
- Not investment advice: Any content produced by this project does not constitute investment, financial, legal, or other professional advice; quoting others' content does not mean agreeing with their views.
- Risk-assumption clause: Use constitutes agreement to all the above terms. If your region or use case does not allow such tools, please stop using and delete this project immediately.
MIT © xiaoyaya191