# Context OS by ContextLM

> Context OS：把资料整理成可分享、可被 Agent 检索的知识库（项目）。本文件列出公开帮助文档的机器可读 Markdown 版本。

## 开始使用

- [认识 Context OS](https://app.contextlm.top/help/getting-started/overview.md): Context OS 是什么、能做什么，以及个人对话与项目两种基本用法。
- [快速上手](https://app.contextlm.top/help/getting-started/quickstart.md): 注册后三分钟完成第一次提问：个人对话直接问，项目模式先传资料再问。
- [个人对话和项目的区别](https://app.contextlm.top/help/getting-started/chat-vs-project.md): 个人对话适合临时提问，项目是可持续积累、可分享的知识容器——按任务性质选择。

## 对话

- [对话附件](https://app.contextlm.top/help/chat/attachments.md): 随提问临时附带的文件：只用于当前这一轮，不进入项目资料；上限为每轮 5 个、单文件 20 MiB、合计提取文字 64 KiB。
- [引用与来源预览](https://app.contextlm.top/help/chat/citations.md): 项目问答中的每条结论都可回溯到资料原文：引用卡标注文件名与页码，点击预览原始片段；已删除资料的引用会标记失效。
- [提问技巧](https://app.contextlm.top/help/chat/prompting.md): 让回答更可靠的提问方法：一次一事、指明资料范围、给出约束与输出格式，并要求模型区分「资料提到」与「模型推断」。

## 项目与资料

- [上传资料](https://app.contextlm.top/help/projects/upload.md): 项目资料的三种加入方式（文件、网页链接、粘贴文本）、支持的格式与大小上限、处理状态含义，以及上传后何时可检索。
- [资料组织建议](https://app.contextlm.top/help/projects/organize.md): 让检索更准的资料写法：用标题层级划分语义边界、一段只讲一件事、表格清单保持原子化、跨章节写明出处。

## 分享

- [访客问答与所有者付费](https://app.contextlm.top/help/sharing/owner-pays.md): 项目分享为公开链接后，访客免登录即可提问，问答产生的模型费用由项目所有者从钱包余额中结算，而不是向访客收费。

## 接入方式

- [云端 API](https://app.contextlm.top/help/api-access.md): 内容托管在云端、查询由平台 Agent 执行的接入方式。在分享中心创建云端密钥，复制 Agent Pack 交给 Cursor / Claude 等外接 Agent；密钥限单项目 index/query 权限，可随时撤销。
- [Agent 快速接入](https://app.contextlm.top/help/api-access/agent-quickstart.md): 三步把 Context OS 项目接进 Agent：创建云端密钥、复制 Agent Pack、跑通第一次检索。本页可直接交给 Agent 阅读执行；内容须留在本机时走本地 MCP。
- [API Access for Agents](https://app.contextlm.top/help/api-access/agents.md): Agent-readable reference for project-scoped access in both tracks — Cloud API (hosted content) and Local MCP (self-hosted instance via context-os-mcp). Agent Pack fields, MCP endpoint, tool catalog, REST surface, error codes.
- [本地 MCP](https://app.contextlm.top/integrations/mcp.md): 内容留在本机、入库与检索都在本地完成的接入方式。你的 Agent 经 context-os-mcp（stdio）连本机 Context OS 实例；只有 Jev 评判与 llm.forward 可能出本机，且可自配密钥。
- [接入 Cursor](https://app.contextlm.top/integrations/cursor.md): 让 Cursor 里的 Agent 检索你的 Context OS 项目：云端项目走远程 MCP 端点 + 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Claude Desktop](https://app.contextlm.top/integrations/claude-desktop.md): 让 Claude 检索你的 Context OS 项目：云端项目走自定义连接器 + 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Trae](https://app.contextlm.top/integrations/trae.md): 让 Trae 里的智能体检索你的 Context OS 项目：云端项目走远程 MCP 端点 + 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 WorkBuddy](https://app.contextlm.top/integrations/workbuddy.md): 让腾讯 WorkBuddy 的 Agent 检索你的 Context OS 项目：云端项目经 mcp-remote 桥接远程端点 + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 Cherry Studio](https://app.contextlm.top/integrations/cherry-studio.md): 让 Cherry Studio 的助手检索你的 Context OS 项目：云端项目走 streamable HTTP 远程服务 + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 ChatWise](https://app.contextlm.top/integrations/chatwise.md): 让 ChatWise 里的助手检索你的 Context OS 项目：云端项目走远程 MCP 服务 + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入通义灵码](https://app.contextlm.top/integrations/lingma.md): 让通义灵码（Lingma / Qoder）的智能体检索你的 Context OS 项目：云端项目经 mcp-remote 桥接远程端点 + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 Kimi Code](https://app.contextlm.top/integrations/kimi-code.md): 让 Kimi Code CLI 检索你的 Context OS 项目：云端项目走 HTTP 远程 MCP + 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 MiniMax Code](https://app.contextlm.top/integrations/minimax-code.md): 让 MiniMax Code（mcode）的 Agent 检索你的 Context OS 项目：云端项目走 streamable-http 远程 MCP + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 Qwen Code](https://app.contextlm.top/integrations/qwen-code.md): 让 Qwen Code 检索你的 Context OS 项目：云端项目走 http 远程 MCP + 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 ZCode](https://app.contextlm.top/integrations/zcode.md): 让 ZCode（智谱 Z.ai 系 IDE）的 Agent 检索你的 Context OS 项目：云端项目走远程 MCP 服务 + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 DeepSeek Harness](https://app.contextlm.top/integrations/deepseek-harness.md): 让 DeepSeek Harness（dsh）的 Agent 检索你的 Context OS 项目：云端项目经 cordis.yml 插件条目走 streamable-http + 云端密钥；本地内容走 context-os-mcp（stdio）。
- [接入 Claude Code](https://app.contextlm.top/integrations/claude-code.md): 让 Claude Code 检索你的 Context OS 项目：云端项目走远程 MCP（HTTP transport + Bearer 云端密钥）；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Windsurf](https://app.contextlm.top/integrations/windsurf.md): 让 Windsurf Cascade 检索你的 Context OS 项目：云端项目在 mcp_config.json 里配 serverUrl + Bearer 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Cline](https://app.contextlm.top/integrations/cline.md): 让 Cline（VS Code 扩展 / CLI）检索你的 Context OS 项目：云端项目配 streamableHttp 远程 server + Bearer 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Codex CLI](https://app.contextlm.top/integrations/codex-cli.md): 让 OpenAI Codex 检索你的 Context OS 项目：云端项目在 config.toml 里配 url + bearer_token_env_var 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。
- [接入 Gemini CLI](https://app.contextlm.top/integrations/gemini-cli.md): 让 Gemini CLI 检索你的 Context OS 项目：云端项目在 settings.json 里配 httpUrl + Bearer 云端密钥；本地内容走 context-os-mcp（stdio）连本机实例。

## 套餐与计费

- [计费方式](https://app.contextlm.top/help/billing/how-billing-works.md): Context OS 的两套独立计量：套餐决定可同时分享的项目名额，钱包余额按实际模型用量计费（jev 与 llm 两类）。访客问答也计入所有者钱包。
- [用量与费用](https://app.contextlm.top/help/billing/usage.md): 用量页怎么看：钱包余额、按模型分列的请求与 token 数、每日花费堆叠图，以及 24 小时/7 天/30 天三个时间档的切换。

## 故障排查

- [资料处理失败或长时间处理中](https://app.contextlm.top/help/troubleshooting/source-failed.md): 资料状态停在「失败/异常」或长时间不变成「就绪」时的排查：加密与扫描件、文件损坏、过大文件与重传方法。
- [回答缺少依据](https://app.contextlm.top/help/troubleshooting/no-evidence.md): 项目提问拿不到带引用的回答时的自查顺序：确认资料已就绪、换用资料中的术语、补足原始数据、检查资料范围勾选。
- [提问被安全策略拦截](https://app.contextlm.top/help/troubleshooting/blocked.md): 提问或回答被安全护栏介入时的表现、常见触发原因，以及如何改写提问继续。
- [余额不足与限速](https://app.contextlm.top/help/troubleshooting/balance.md): 钱包余额不足或触发平台保护限速时的表现与处理：充值入口、套餐区别、限速是滚动窗口保护而非扣费。

## 参考

- [常见问题](https://app.contextlm.top/help/faq.md): Context OS 常见问题：产品定位、项目与分享、计费与接入三组问答；细节问题见各专题文档。
- [选型对比](https://app.contextlm.top/help/compare.md): Context OS 与笔记内置 AI、通用 RAG 自建栈的类别级对照：核心对象、入库问答、对外分享、外接 Agent 与计费模式的差异。
- [检索范式对比：向量 / 图谱 / PageIndex / Jev 编译式](https://app.contextlm.top/compare/retrieval-paradigms.md): 四种检索范式的底物、判定方式、失败语义与成本结构对照——Context OS 采用 Jev 编译式检索：入库时把每段内容编译成可读的分面坐标，查询期路由导航而非相似度召回。
- [Context OS 与 Notion AI](https://app.contextlm.top/compare/notion-ai.md): 按固定维度对照：核心对象、入库问答、对外分享、外接 Agent、计费模型。Notion 是文档协作平台加内嵌 AI，Context OS 是可分享、可被外接 Agent 检索的项目知识库。
- [Context OS 与 NotebookLM](https://app.contextlm.top/compare/notebooklm.md): 按固定维度对照：核心对象、入库问答、对外分享、外接 Agent、计费模型。NotebookLM 是围绕笔记本的资料研读助手，Context OS 是可分享、可被外接 Agent 检索的项目知识库。
- [Context OS 与腾讯 ima](https://app.contextlm.top/compare/ima.md): 按固定维度对照：核心对象、入库问答、对外分享、外接 Agent、计费模型。ima 是腾讯系个人知识库产品（有官方 OpenAPI），Context OS 强调项目边界、访客免登录问答与双轨 Agent 接入。
- [Context OS 与 Obsidian + AI 插件](https://app.contextlm.top/compare/obsidian-ai.md): 按固定维度对照：核心对象、入库问答、对外分享、外接 Agent、计费模型。Obsidian 是本地 Markdown 库加社区 AI 插件，Context OS 是可直接对外分享问答、官方双轨 Agent 接入的项目知识库。
- [场景：论文/文献库问答](https://app.contextlm.top/scenarios/paper-library.md): 把一批论文或文献收成项目，自己和访客都能直接提问；回答带出处证据，可给外接 Agent 复用同一库。
- [场景：对外公开知识库问答](https://app.contextlm.top/scenarios/public-kb-qa.md): 把产品手册、FAQ、规范文档收成项目并开公开分享，访客免登录直接提问；所有者付费、配额可控。
- [场景：给 Agent 接入的可检索知识库](https://app.contextlm.top/scenarios/agent-kb.md): 把项目文档/代码库说明收成项目，外接 Agent 经云端 API 或本地 MCP 直接检索；项目密钥即最小权限边界。

## English

### Getting started

- [What is Context OS](https://app.contextlm.top/en/help/getting-started/overview.md): What Context OS is and does, plus the two basic modes — standalone chats and projects.
- [Quickstart](https://app.contextlm.top/en/help/getting-started/quickstart.md): From sign-up to your first answer in three minutes — ask directly in a personal chat, or upload sources into a project first.
- [Standalone chat vs projects](https://app.contextlm.top/en/help/getting-started/chat-vs-project.md): Standalone chats are for one-off questions; projects are durable, shareable knowledge containers — pick by the nature of the task.

### Chat

- [Chat attachments](https://app.contextlm.top/en/help/chat/attachments.md): Files attached ad hoc to a single turn — used only for that turn, never added to project sources. Limits: 5 files per turn, 20 MiB each, 64 KiB total extracted text.
- [Citations and source preview](https://app.contextlm.top/en/help/chat/citations.md): Every claim in a project answer traces back to a source — citation cards show the file name and page, click to preview the original passage; citations of deleted sources are marked expired.
- [Prompting tips](https://app.contextlm.top/en/help/chat/prompting.md): How to ask more reliably — one thing per turn, name the source scope, give constraints and an output format, and ask the model to separate "stated in sources" from "model inference".

### Projects & sources

- [Uploading sources](https://app.contextlm.top/en/help/projects/upload.md): Three ways to add project sources (files, web URLs, pasted text), supported formats and size limits, what processing states mean, and when an upload becomes retrievable.
- [Organizing sources](https://app.contextlm.top/en/help/projects/organize.md): Writing practices that make retrieval sharper — use heading levels as semantic boundaries, one topic per paragraph, atomic tables and lists, and name cross-references explicitly.

### Sharing

- [Visitor Q&A and owner-pays billing](https://app.contextlm.top/en/help/sharing/owner-pays.md): Once a project is shared as a public link, visitors can ask without logging in; the model cost of their questions settles to the project owner's wallet — visitors are never charged.

### Access tracks

- [Cloud API](https://app.contextlm.top/en/help/api-access.md): The hosted access track — content lives on Context OS servers and queries run on the platform agent. Create a cloud key in the share center, copy the Agent Pack into Cursor / Claude; keys are scoped to one project with index/query permissions and can be revoked anytime.
- [Agent quickstart](https://app.contextlm.top/en/help/api-access/agent-quickstart.md): Connect a Context OS project to your agent in three steps — mint a cloud key, copy the Agent Pack, run the first query. Agents can read and execute this page directly; use Local MCP when content must stay on-device.
- [API Access for Agents](https://app.contextlm.top/en/help/api-access/agents.md): Agent-readable reference for project-scoped access in both tracks — Cloud API (hosted content) and Local MCP (self-hosted instance via context-os-mcp). Agent Pack fields, MCP endpoint, tool catalog, REST surface, error codes.
- [Local MCP](https://app.contextlm.top/en/integrations/mcp.md): The integration track where content stays on your machine and ingestion + retrieval run locally. Your agent connects through context-os-mcp (stdio) to a local Context OS instance; only Jev evaluation and llm.forward may leave the machine, and you can bring your own credentials.
- [Connect Cursor](https://app.contextlm.top/en/integrations/cursor.md): Let the agent in Cursor query your Context OS project: cloud projects connect via the remote MCP endpoint + cloud key; local content uses context-os-mcp (stdio) against a local instance.
- [Connect Claude Desktop](https://app.contextlm.top/en/integrations/claude-desktop.md): Let Claude query your Context OS project: cloud projects connect via a custom connector + cloud key; local content uses context-os-mcp (stdio) against a local instance.
- [Connect Trae](https://app.contextlm.top/en/integrations/trae.md): Let agents in Trae query your Context OS project: cloud projects connect via the remote MCP endpoint + cloud key; local content uses context-os-mcp (stdio) against a local instance.
- [Connect WorkBuddy](https://app.contextlm.top/en/integrations/workbuddy.md): Let Tencent WorkBuddy's agent query your Context OS project: cloud projects connect through an mcp-remote bridge to the remote endpoint + cloud key; local content uses context-os-mcp (stdio).
- [Connect Cherry Studio](https://app.contextlm.top/en/integrations/cherry-studio.md): Let assistants in Cherry Studio query your Context OS project: cloud projects connect via a streamable HTTP remote server + cloud key; local content uses context-os-mcp (stdio).
- [Connect ChatWise](https://app.contextlm.top/en/integrations/chatwise.md): Let assistants in ChatWise query your Context OS project: cloud projects connect via a remote MCP server + cloud key; local content uses context-os-mcp (stdio).
- [Connect Lingma (Qoder)](https://app.contextlm.top/en/integrations/lingma.md): Let the agent in Lingma / Qoder query your Context OS project: cloud projects connect through an mcp-remote bridge to the remote endpoint + cloud key; local content uses context-os-mcp (stdio).
- [Connect Kimi Code](https://app.contextlm.top/en/integrations/kimi-code.md): Let Kimi Code CLI query your Context OS project: cloud projects connect via the HTTP remote MCP endpoint + cloud key; local content uses context-os-mcp (stdio) against a local instance.
- [Connect MiniMax Code](https://app.contextlm.top/en/integrations/minimax-code.md): Let the agent in MiniMax Code (mcode) query your Context OS project: cloud projects connect via a streamable-http remote MCP server + cloud key; local content uses context-os-mcp (stdio).
- [Connect Qwen Code](https://app.contextlm.top/en/integrations/qwen-code.md): Let Qwen Code query your Context OS project: cloud projects connect via the http remote MCP endpoint + cloud key; local content uses context-os-mcp (stdio) against a local instance.
- [Connect ZCode](https://app.contextlm.top/en/integrations/zcode.md): Let the agent in ZCode (Zhipu's Z.ai-family IDE) query your Context OS project: cloud projects connect via a remote MCP server + cloud key; local content uses context-os-mcp (stdio).
- [Connect DeepSeek Harness](https://app.contextlm.top/en/integrations/deepseek-harness.md): Let the agent in DeepSeek Harness (dsh) query your Context OS project: cloud projects connect through a cordis.yml plugin entry over streamable-http + cloud key; local content uses context-os-mcp (stdio).
- [Connect Claude Code](https://app.contextlm.top/en/integrations/claude-code.md): Let Claude Code query your Context OS project — cloud projects via remote MCP (HTTP transport + Bearer cloud key); local content via context-os-mcp (stdio) to your local instance.
- [Connect Windsurf](https://app.contextlm.top/en/integrations/windsurf.md): Let Windsurf Cascade query your Context OS project — cloud projects via serverUrl + Bearer cloud key in mcp_config.json; local content via context-os-mcp (stdio) to your local instance.
- [Connect Cline](https://app.contextlm.top/en/integrations/cline.md): Let Cline (VS Code extension / CLI) query your Context OS project — cloud projects via a streamableHttp remote server + Bearer cloud key; local content via context-os-mcp (stdio) to your local instance.
- [Connect Codex CLI](https://app.contextlm.top/en/integrations/codex-cli.md): Let OpenAI Codex query your Context OS project — cloud projects via url + bearer_token_env_var in config.toml; local content via context-os-mcp (stdio) to your local instance.
- [Connect Gemini CLI](https://app.contextlm.top/en/integrations/gemini-cli.md): Let Gemini CLI query your Context OS project — cloud projects via httpUrl + Bearer cloud key in settings.json; local content via context-os-mcp (stdio) to your local instance.

### Plans & billing

- [How billing works](https://app.contextlm.top/en/help/billing/how-billing-works.md): Context OS has two independent meters — the plan controls how many projects you can share at once, and wallet balance bills actual model usage (jev and llm classes). Visitor Q&A also bills to the owner's wallet.
- [Usage and spend](https://app.contextlm.top/en/help/billing/usage.md): How to read the usage page — wallet balance, per-model request and token counts, a daily stacked-spend chart, and the 24-hour / 7-day / 30-day range switch.

### Troubleshooting

- [Source failed or stuck processing](https://app.contextlm.top/en/help/troubleshooting/source-failed.md): What to check when a source stays at "failed/error" or never reaches "ready" — encryption and scans, corrupt files, oversize files, and how to re-upload.
- [Answer lacks evidence](https://app.contextlm.top/en/help/troubleshooting/no-evidence.md): Self-check order when a project question returns an uncited answer — confirm sources are ready, reuse the sources' own terminology, supply raw data, and check the source selection.
- [Question blocked by safety policy](https://app.contextlm.top/en/help/troubleshooting/blocked.md): What it looks like when a question or answer triggers safety intervention, the common causes, and how to rephrase and continue.
- [Insufficient balance and rate limits](https://app.contextlm.top/en/help/troubleshooting/balance.md): What happens when the wallet runs low or platform rate-limit protection trips — where to top up, how plans differ, and why limits are rolling-window protection rather than charges.

### Reference

- [FAQ](https://app.contextlm.top/en/help/faq.md): Frequently asked questions about Context OS — what it is, projects and sharing, billing and agent access. Details live in the topic guides.
- [Compare](https://app.contextlm.top/en/help/compare.md): Context OS vs notes-with-AI apps vs general RAG stacks — a category-level comparison of core object, ingest and Q&A, external sharing, agent access and billing models.
- ["Retrieval paradigms compared: vectors / knowledge graphs / PageIndex / Jev compiled"](https://app.contextlm.top/en/compare/retrieval-paradigms.md): Four retrieval paradigms side by side — substrate, decision mechanism, failure semantics, and cost structure. Context OS uses Jev compiled retrieval: at ingest time every chunk is compiled into a readable faceted coordinate; at query time a router navigates rather than similarity-recalling.
- [Context OS vs Notion AI](https://app.contextlm.top/en/compare/notion-ai.md): Fixed-dimension comparison — core object, ingest & Q&A, external sharing, external agents, billing model. Notion is a docs collaboration platform with embedded AI; Context OS is a project knowledge base that is shareable and queryable by external agents.
- [Context OS vs NotebookLM](https://app.contextlm.top/en/compare/notebooklm.md): Fixed-dimension comparison — core object, ingest & Q&A, external sharing, external agents, billing model. NotebookLM is a per-notebook source-study assistant; Context OS is a shareable project knowledge base queryable by external agents.
- [Context OS vs Tencent ima](https://app.contextlm.top/en/compare/ima.md): Fixed-dimension comparison — core object, ingest & Q&A, external sharing, external agents, billing model. ima is Tencent's personal knowledge-base product (with an official OpenAPI); Context OS emphasizes project boundaries, guest login-free Q&A, and a dual-track agent surface.
- [Context OS vs Obsidian + AI plugins](https://app.contextlm.top/en/compare/obsidian-ai.md): Fixed-dimension comparison — core object, ingest & Q&A, external sharing, external agents, billing model. Obsidian is a local Markdown vault plus community AI plugins; Context OS is a project knowledge base with direct guest Q&A sharing and an official dual-track agent surface.
- ["Scenario: paper / literature library Q&A"](https://app.contextlm.top/en/scenarios/paper-library.md): Collect a batch of papers or literature into a project so you and your visitors can ask directly; answers carry cited evidence, and external agents can query the same base.
- ["Scenario: public knowledge-base Q&A"](https://app.contextlm.top/en/scenarios/public-kb-qa.md): Collect product manuals, FAQs, and regulatory documents into a project and open a public share — visitors ask directly without logging in; the owner pays and quotas are controllable.
- ["Scenario: a searchable knowledge base for agents"](https://app.contextlm.top/en/scenarios/agent-kb.md): Collect project docs or codebase references into a project and let external agents query it directly via the Cloud API or local MCP; the project key is the least-privilege boundary.