本地优先 Web 应用Local-first web app· 2026
Planbox
AI 活动机会管理器AI opportunity manager
把散落在微信、小红书和截图里的活动信息，整理成可决策、可提醒的机会看板。A local-first workspace for collecting, comparing, and acting on event or opportunity signals.
案例摘要Case summary
目标用户Audience
经常收集 AI 活动、黑客松、训练营信息的学生、开发者和产品经理。Students, developers, and product managers who frequently collect AI events, hackathons, and bootcamp information.
我的角色Role
负责 PRD、MVP 范围、人工确认节点和核心数据流的产品决策；全部代码由 AI 编程工具完成。Owned the PRD, MVP scope, human-review checkpoints, and core data-flow decisions; all code was produced with AI coding tools.
交付结果Outcome
形成“输入活动信息 -> 保存原始证据 -> AI 抽取 -> 人工确认 -> 入库”的最小验证链路。Built the minimum validation chain: input event info → save raw evidence → AI extraction → human confirmation → store.
产品洞察Product insight
收藏不等于行动。第一版不追求自动爬取，而是先验证“信息能否可靠变成行动”，所以把原始证据、AI 抽取和人工确认解耦。Bookmarking is not acting. The first version does not chase automatic scraping; it first validates whether information can reliably become action — so raw evidence, AI extraction, and human confirmation are decoupled.
项目简介Overview
Planbox 是本地优先的活动机会管理工具，支持保存原文、AI 结构化抽取、人工确认、活动状态管理和日历导出思路。Planbox is a local-first event opportunity management tool that supports saving the original text, AI structured extraction, human confirmation, event status management, and calendar-export ideas.
技术栈Tech stack
结果证据Evidence
docs/PRD.md 和 docs/SOLUTION.md 先定义边界、数据模型、风险和里程碑，再进入实现。docs/PRD.md and docs/SOLUTION.md define the boundary, data model, risks, and milestones before implementation.
本地前端已跑通 Inbox、Review、Activities、Brief 四个主视图。The local frontend runs four main views: Inbox, Review, Activities, and Brief.
后端使用 FastAPI + uv + SQLite 起步，保留原文证据和人工确认流程。The backend starts with FastAPI + uv + SQLite, keeping raw evidence and a human-confirmation flow.
项目截图Screenshots
关键决策Key decisions
第一版直接做微信/小红书自动爬取，会被登录态、风控和平台限制拖死。Directly building WeChat/Xiaohongshu auto-scraping in v1 would be dragged down by login states, risk control, and platform limits.
先支持用户主动粘贴文本、链接和截图，把自动采集延后为可选适配器。Start by letting users paste text, links, and screenshots; defer automatic collection to optional adapters.
AI 可能误读时间、地点和报名链接。The AI can misread times, locations, and signup links.
抽取结果必须进入 Review Queue，并保留字段级证据和置信度。Extraction results must enter a Review Queue, keeping field-level evidence and confidence.
想看完整的代码和实现细节？Want the full code and implementation details?
在 GitHub 上查看View on GitHub