Second Brain
A persistent memory layer for AI tools, deploy to your own Cloudflare account in 5 minutes, letting Claude, ChatGPT, Cursor and all AI tools share the same memory.
In-depth Report
-
Second Brain is an open source cross-AI tool persistence memory layer application that can be deployed on the Mac/Windows desktop in 5 minutes. It runs on the user's own Cloudflare account and provides unified, searchable, and controllable persistent memory for mainstream AI tools such as Claude, ChatGPT, and Cursor, allowing project context, decision records, and personal knowledge to seamlessly continue between different AI sessions. The product adopts the MIT open source license and is free for personal use. The data is 100% owned by the user. It ranked third on Product Hunt that day (276 upvotes).
-
The current AI tool ecosystem faces a core contradiction: every session is a "whiteboard." A series of key contexts such as project decisions, code specifications, personal preferences, and technology selection when users collaborate using multiple tools such as Claude, ChatGPT, Cursor, and Windsurf are completely lost after closing chat. You can only paste the context over and over in subsequent sessions, wasting 10-15 minutes each time re-establishing the session context. Each memory solution on the market has limitations: the built-in memory function of each platform is a black box, cannot be searched, reviewed or exported, triggering is uncontrollable, and recall accuracy is low; mem0 has usage billing; GitHub Models are limited to coding scenarios. Second Brain provides a free, decentralized, cross-all tool solution through the MCP (Model Context Protocol) standard protocol.
-
core memory skills AI Synthesis: Ask "What is the pricing strategy?" and get a comprehensive answer directly instead of a pile of original items. The underlying support memory can be expanded and viewed. Semantic search (Search by Meaning): Searching for "conversion issues" can search for records of "users losing during checkout" - it does not rely on keyword matching and does not require manual maintenance of tags. Importance Scoring: Each memory is scored by AI when it is saved, and content that is frequently reviewed will naturally increase its ranking over time. Intelligent management function Contradiction Detection: Automatically mark and make decisions when the new memory conflicts with the semantics of existing records. The newer version shall prevail. Smart Merge: Approximate memories with a similarity of 0.85-0.95 are judged by LLM - keep both/replace/merge into a single coherent record. Hard Duplicate Block: Exact or approximate matching (≥0.95 similarity) blocks directly without LLM call. Ecosystem integration Obsidian plug-in: allows the local note library to be read by AI tools, and supports filtering the synchronization range by note tags. CLI tool: "brain remember cat notes.md" in the terminal is written directly into the memory, and "brain recall "state library decision"" can be searched. iOS shortcut command: Record your thoughts with your voice while walking and save them to memory with one click. Web dashboard: Browse by date, filter by tag, edit memories in place, full-text JSON/Markdown export. Mailbox/Calendar Integration: Automatically filter newsletters and promotional emails to only see "real content"; calendar events are converted to a timeline for querying. Notion synchronization: Turn on Nightly automatic synchronization, and the pages shared after connecting to the Notion workspace will automatically enter the memory library. Technical advantages Runs within the user's Cloudflare account (D1 + Vectorize + Workers), and the AI model call is also completed under the same account. Adopts MCP protocol standard and is compatible with all AI clients that support this protocol. The free plan covers personal use (Cloudflare Free Plan) and no credit card is required.Open source MIT, you can fork/audit/build your own version and is not restricted by upstream manufacturers.
-
Developer scenario: Claude Code does architectural design, Cursor writes code, ChatGPT does brainstorming - all contexts are shared between tools. The rationale for technology selections, historical decisions, and legacy design constraints are automatically presented across sessions. Project context is automatically accumulated across multiple AI sessions, eliminating the need for repeated pasting. Knowledge worker scenario: Store customer contact information, meeting minutes, and communication history, and then directly ask "What did this customer say last time?" Article ideas, structural design, and revision history can be accumulated across time periods. Study notes, reading excerpts, and course summaries can be intelligently retrieved in AI chat after being synchronized through Obsidian. Team collaboration scenario: Multiple AI clients (Claude, ChatGPT, Cursor, etc.) read the same memory library. When you come back from vacation, just ask "Where did I go with my project?" and the memory will be restored automatically.
-
Download the desktop app: Mac or Windows version Cloudflare automatically configures the D1 database, Vectorize indexes, and workers with zero configuration. Connection tool: Paste the paired Worker URL into the configuration of any client that supports MCP, such as Claude Desktop, Claude Code, ChatGPT, Cursor, Windsurf, etc. One URL for all tools. Start accumulating: Daily work naturally fills the memory, and relevant content automatically emerges in subsequent conversations. The longer the time, the greater the accumulated value.
-
Compared with similar products (mem0, GitHub Models, Second Brain), the core differences of Second Brain are: self-hosted own infrastructure, free for personal use, no usage billing, covering all AI tools instead of just coding scenarios, the same memory on all devices, AI models run within their own accounts, supports Obsidian/browser/iOS saving, and has conflict detection and intelligent merging functions.
-
MIT License, core functionality is completely free, and runs on Cloudflare’s free plan, no credit card required, and no usage billing. Data stays within their own Cloudflare account and does not flow from the user to third parties. After forking the code, it can run independently from the upstream.
-
The core positioning is the "shared memory layer of AI tools" - not a note-taking application, not a chatbot, but an infrastructure that allows all AI assistants to have the same long-term memory. The difference is: data self-sustaining (own Cloudflare account), cross-tool (MCP standard protocol), open source MIT (no lock-in risk).
User Reviews
-
OctoOps74—部署了 Second Brain，最直接的感受就是再也不用在 Cursor 和 Claude 之间反复粘贴上下文了。项目架构记录一次，两边都能调出来，省心不少。 -
CeliaMartínez—Cloudflare 免费套餐就能跑，自己控制数据，这点比 Mem0 那种按条收费的良心太多了。不过部署环节对新手还是有门槛的，至少得懂 Wrangler 是怎么用的。 -
于浩刚—部署很丝滑，一键 Cloudflare 就完事了。现在 Claude Desktop 和 Cursor 共享一个记忆库，感觉自己终于不用每次开新会话都报一遍家底了。 -
MichelleSullivan_7158—试了一周，语义搜索确实靠谱。存的笔记里写的是『用户在支付环节流失』，问系统『转化问题』居然能匹配到。第一次觉得 AI 工具间的记忆不再是黑盒了。 -
Paul469—矛盾和逻辑冲突检测这个功能有意思。今天存了一个『团队倾向于用 Postgres』，明天下了一个『考虑迁移到 SQLite』的系统自动标注冲突并询问是要覆盖还是保留，挺聪明的。 -
Helen.Baker_7—Obsidian 插件好评。我几千条笔记一键同步过去，AI 直接能读取，再也不用手动复制粘贴了。不过首次同步有点慢，建议晚上跑。 -
TheAnjaRodriguez_2024—用了两周，最大的感受是『大脑不再清零了』。之前用 AI 最大的痛点是每次新会话都是白板，现在至少能找到上次的决策记录，不用靠记忆硬扛了。 -
MJames_Max—说句实话，这东西的核心思路没错，但要做好心理准备——要让它真正有用，你得养成存记忆的习惯。不是装上了就自动变聪明的。 -
ChristineJohnson_X—项目决策记录这个场景太绝了。做技术选型时在 Claude 里讨论得出的结论，下次在 Cursor 里写代码时它自动就 recall 出来了。再也不用回头看聊天记录了，回不去了。 -
Arthur.Cox_2020—开源 MIT 协议还是很有诚意的，想怎么改怎么改。我把它改了改接入了自己的几个私有工具，数据全在自家 Cloudflare 账户里，踏实。 -
NajaChristiansen—从 Mem0 切过来的，一样是自托管，但 Second Brain 免费还不限量，两边效果差不多。早切早省钱。 -
Pamela_HughesZ—刚部署完，提个小建议：dashboard 的搜索如果能加个按标签过滤就更好了。现在只能全文搜，东西一多有点大海捞针的感觉。 -
KRussell_66—V2 的记忆图谱很不错。相关记忆自动关联起来之后，问问题出来的信息比单条 recall 丰富得多。问了『上次关于定价的讨论』，结果把会议纪要、邮件往来、最终决定全串起来了。 -
TGomez_9915—Notion 同步功能对我这种重度用户来说很关键。不过 Nightly 同步有时候不太触发，得手动点 Sync now。希望后面版本能优化一下。 -
Angela.Davis520—yyds！装好之后跟 Claude 说『记住我当前项目的技术栈』，然后换到 ChatGPT 问相同的问题，它真能答出来。跨工具记忆确实不是概念了。 -
Roxin764—适合开发者，不太适合纯小白。虽然官方说入门两分钟，但如果你不懂 MCP、不会配 Cloudflare，还是得有个人帮忙指导。 -
RReedQ—Product Hunt 上那个版本我就在用了，当时就觉得概念很酷。V2 加了 Notion 同步和知识图谱后实用性强了很多，真的开始像『第二大脑』了。 -
Katherine_Allen_2021—跟同类的 Mem0 比各有优劣。Mem0 的召回准确率略高一点，但 Second Brain 免费且全量，Cloudflare 的延迟也低，日常用完全够了。 -
AFlores007—iOS 快捷指令太好用了。走路时突然想到什么，掏出手机喊一句就记进去了。回头在 Claude 里问『上周我有什么想法』，它都能列出来。 -
DrKiaraLaurent_dev—最近把工作中所有关于收费策略的思考都存进去了，今天问了一次系统，给出的答案清晰有条理。AI 终于开始积累知识而非每次都从零开始了。 -
Victoria.Collins—部署前看了好几遍 README，照着一键部署按钮点了就成功了。不过建议大家设置 AUTH_TOKEN 的时候用 openssl 生成个强密码，别用太简单的。 -
LeviKin_g—这东西的正确用法是当基础设施，不是当应用。把它当记忆 API 来用，任何 MCP 客户端都能接进来，潜力比单工具内置记忆大得多。 -
rEBECCAmURPHY—有个问题：当我几个月前的决策后来被推翻了，它怎么处理？目前好像没有优雅的版本管理机制。新记忆直接覆盖旧的，但有时候我想保留旧记录做审计用。 -
realمحمدطاهامحمدخان—如果你平时只用单一 AI 工具，可能感觉不到它的价值。但如果你像我一样 Claude/Cursor/ChatGPT 轮流用的，这玩意就是刚需。 -
HeatherPerez_2022—免费且开源，数据由自己掌控——就冲这两点就该试试。别让 AI 厂商把你锁在他们的记忆黑盒里。 -
琥珀_12—用了快一个月，记忆库大概有 200 多条记录了。回忆的准确率随着数据增多反而越来越高，说明它的重要性评分机制确实在起作用。 -
HaroldJohnsonK98—官网设计也挺有品味的，两个工具对比页写得清楚直白。感觉开发者对自己的产品定位很清晰：不是笔记应用，是 AI 的记忆层。 -
Judith.DavisII7—CLI 工具挺好用的，brain remember 接管道输出，brain recall 查历史，比打开网页操作快多了。但初次配置 CLI 的那个 token 路径找了半天。 -
书生583—之前一直用 Cursor 的 Composer 写代码，上下文一长就卡。现在把关键决策存 Second Brain，每次开始新会话它自动 recall，体感上对话质量提升很明显。 -
ProtocolPal61488—期待后续能支持更多数据源的自动接入，比如 Slack 消息归档、GitHub Issue 自动同步，这样记忆库就真能跑在真实工作流上了。