为AI代理提供结构化、版本化上下文。Context Nest将知识组织为markdown文档，存储在保险库中，支持版本控制、完整性验证和查询语言——可通过CLI或......
概览
Context Nest 将分散的知识——你的仓库、文档、Slack 帖子、部落知识——转化为一个结构化、可查询的大脑，供你的 AI 代理使用。 这和把笔记放进Obsidian并用LLM指向它们的本能是一样的，但这个模式并没有给你四样东西： - 带有关系和选择语法的类型节点。代理通过图表导航，而非平面文件夹。- 预先消化为摘要和链接枢纽文档，使下一次会话读取~500个相关上下文标记，而非填充5万个原始文件标记。- 导出给队友，或作为付费包发布到PromptOwl市场，供他人查询。- 每项变更都经过版本控制和哈希链。完整的审计轨迹，批准准备，可精确到字节。同一仓库在十分钟内引导一位开发者上线，当那天通过SOC 2审查。
概述（AI 翻译）
为你的AI代理打造一个结构化的第二大脑。单人开始，安全扩展。
作者：PromptOwl |网站 |白皮书 |规范 |Discord
Context Nest 将分散的知识——你的仓库、文档、Slack 帖子、部落知识——转化为一个结构化、可查询的大脑，供你的 AI 代理使用。
这和把笔记放进Obsidian并用LLM指向它们的本能是一样的，但这个模式并没有给你四样东西：
- 结构。 带有关系和选择语法的类型节点。代理通过图来导航，而不是平面文件夹。
- ~100×更便宜的代理会话。 预先消化成摘要和链接集线器文档，因此下一次会话读取 ~500 个相关上下文标记，而不是塞入 50k 个原始文件标记。
- 共享路径。 导出给队友，或作为付费包发布到 PromptOwl 市场，供他人查询。
- 被治理，不是合规附加。 每一次变更都经过版本控制和哈希链。完整的审计轨迹，批准准备，可审计到字节。当那天到来时，那个在十分钟内就让一位开发者上线的保险库通过了 SOC 2 审核。
本页由 AI 翻译，技术细节请以原文为准。
README
以下为原文 README 节选，完整内容请查看原文。
Context Nest
A structured second brain for your AI agents. Start solo, scale safely.
by PromptOwl | Website | Whitepaper | Specification | Discord
Context Nest turns scattered knowledge — your repos, docs, Slack threads, tribal know-how — into a structured, queryable brain your AI agents can use.
It’s the same instinct as dumping your notes into Obsidian and pointing an LLM at them, with four things that pattern doesn’t give you:
- Structure. Typed nodes with relationships and a selector grammar. Agents navigate a graph, not a flat folder.
- ~100× cheaper agent sessions. Pre-digested into summaries and linked hub documents, so the next session reads ~500 tokens of relevant context instead of stuffing 50k tokens of raw files.
- A sharing path. Export to a teammate, or publish to the PromptOwl marketplace as a paid pack others can query.
- Governed, not a compliance bolt-on. Every change is versioned and hash-chained. Full audit trail, approval-ready, auditable down to the byte. The same vault that onboarded one developer in ten minutes passes a SOC 2 review when that day comes.
Works the same for a solo dev’s second brain, a team’s living onboarding doc, or an enterprise’s safe shared brain — one CLI, one file format, one vault. Start solo; scale when you need to.
Quick Start
npm install -g @promptowl/contextnest-cli
ctx
The first time you run ctx it walks you through setup: a local vault, a team server, or both. After that, ctx just shows the command list. Prefer to script it? ctx init --starter developer still works.
Getting started is one question: what are you trying to capture? Point your agent at a codebase, a folder of docs, an old wiki, or just tell it what’s in your head. It’ll build the first usable version in ten minutes and get denser every time you come back.
See all starters: ctx init --list-starters
What the install puts on your disk
Two packages, no nesting, and no install scripts — nothing of ours executes when you install. The MCP server installs zero dependencies. For a lean install without terminal colour:
npm install -g @promptowl/contextnest-cli --omit=optional
That leaves exactly one package. The rest of the code is compiled into the published bundle rather than resolved from npm, so the install is deterministic — and every bundled package is listed with its version and licence alongside the installed ones in DEPENDENCIES.md. CI regenerates that file on every run and fails on drift, and uploads the machine-readable graph as the dependency-graph build artifact.
What leaves your machine
Nothing, unless you run a command that names a destination or opt in:
| Trigger | Destination | Sent | Default |
|---|
安装
docker run -i --rm contextnest-mcp # demo vault baked into the image配置
{
"mcpServers": {
"contextnest": {
"command": "node",
"args": ["/absolute/path/to/packages/mcp-server/dist/index.js"],
"env": {
"CONTEXTNEST_VAULT_PATH": "/path/to/your/vault"
}
}
}
}