kb-memory 2.1.0__tar.gz

kb_memory-2.1.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 little-fishy
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

kb_memory-2.1.0/PKG-INFO

@@ -0,0 +1,415 @@
+Metadata-Version: 2.4
+Name: kb-memory
+Version: 2.1.0
+Summary: Local-first agent memory & knowledge service: hybrid retrieval + memory governance (decay / dedup / freshness), REST + MCP dual protocol, fully usable offline without an LLM
+Author: fish827-08
+License-Expression: Apache-2.0
+Project-URL: Homepage, https://github.com/fish827-08/rag-kb
+Project-URL: Repository, https://github.com/fish827-08/rag-kb
+Keywords: mcp,memory,agent,rag,knowledge-base,local-first
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: fastapi<1.0,>=0.110
+Requires-Dist: uvicorn[standard]<1.0,>=0.29
+Requires-Dist: mcp<3.0,>=2.0
+Requires-Dist: chromadb<2.0,>=1.5
+Requires-Dist: rank_bm25<0.3,>=0.2.2
+Requires-Dist: jieba<0.43,>=0.42
+Requires-Dist: sentence-transformers<7.0,>=3.0
+Requires-Dist: transformers<6,>=4.40
+Requires-Dist: httpx<1.0,>=0.27
+Requires-Dist: langchain-text-splitters<2.0,>=0.3
+Requires-Dist: pydantic-settings<3.0,>=2.0
+Requires-Dist: typer<1.0,>=0.12
+Requires-Dist: rich<16.0,>=13.0
+Provides-Extra: ingest
+Requires-Dist: pypdf<7.0,>=4.0; extra == "ingest"
+Requires-Dist: python-docx<2.0,>=1.1; extra == "ingest"
+Requires-Dist: markitdown<0.2,>=0.1; extra == "ingest"
+Requires-Dist: trafilatura<3.0,>=1.8; extra == "ingest"
+Requires-Dist: lxml_html_clean<1.0,>=0.4; extra == "ingest"
+Requires-Dist: watchdog<7.0,>=4.0; extra == "ingest"
+Provides-Extra: llm
+Requires-Dist: openai<4.0,>=1.0; extra == "llm"
+Provides-Extra: full
+Requires-Dist: pypdf<7.0,>=4.0; extra == "full"
+Requires-Dist: python-docx<2.0,>=1.1; extra == "full"
+Requires-Dist: markitdown<0.2,>=0.1; extra == "full"
+Requires-Dist: trafilatura<3.0,>=1.8; extra == "full"
+Requires-Dist: lxml_html_clean<1.0,>=0.4; extra == "full"
+Requires-Dist: watchdog<7.0,>=4.0; extra == "full"
+Requires-Dist: openai<4.0,>=1.0; extra == "full"
+Dynamic: license-file
+
+# rag-kb — 本地 Agent 记忆与知识服务
+
+<!-- mcp-name: io.github.fish827-08/kb-memory -->
+
+[![License: Apache-2.0](https://img.shields.io/badge/License-Apache--2.0-blue.svg)](LICENSE)
+[![Glama MCP score](https://glama.ai/mcp/servers/fish827-08/rag-kb/badges/score.svg)](https://glama.ai/mcp/servers/fish827-08/rag-kb)
+[English](README_EN.md) | 中文
+
+**本地优先、完全离线可用的 Agent 记忆与知识服务**——REST + MCP 双协议，混合检索（向量 + BM25 归一化加权融合），文档/网页入库与 RAG 问答；无 LLM 时存取与检索完整可用。
+
+<!-- GitHub Topics: mcp, memory-service, rag, ai-agent, knowledge-base, local-first, llm, claude-code, hybrid-search, embedding -->
+
+**kb** 是本地优先、完全免费的 Agent 记忆与知识服务（v2.1.0 · 自动化回归全绿，待人工验收）。
+
+> 开源协议：[Apache-2.0](LICENSE)（含专利授权，可商用）。
+
+## kb — 本地优先的 Agent 记忆与知识服务
+
+Windows 单进程常驻（`python -m kb serve`），REST + MCP 双协议，向 Claude Code /
+Cursor / TraeWork / 自建 Agent 提供记忆写入、文档与网页入库、混合检索
+（向量 + BM25 归一化加权融合）与 RAG 问答。
+
+**无 LLM 时存取与检索完整可用**——记忆写入、文档入库、混合检索不依赖任何大模型；
+配置本地 Ollama 或云端 API 后，`/ask` 问答能力自动启用。
+
+## 核心特性
+
+- **单进程常驻**：一个 `python -m kb serve` 同时提供 REST API 与 MCP 端点，无需额外组件
+- **混合检索**：向量检索 + BM25 关键词检索，归一化加权融合排序（每路 min-max 归一化后按生效路数均权，N32/N33），中文分词友好
+  - 嵌入模型：**默认 `BAAI/bge-small-zh-v1.5`（轻量，约 92 MB / 512 维）**——开箱成本低、CPU 上也快；需要更强语义理解可换 `BAAI/bge-m3`（约 2.2 GB / 1024 维），见「[嵌入模型怎么选](#嵌入模型怎么选)」
+  - 可选精排：`KB_RERANK_ENABLED=true` 启用 bge-reranker-v2-m3 交叉重排（默认关）
+  - 可选三路：`KB_SPARSE_ENABLED=true` 启用 BGE-M3 稀疏向量第三路（默认关，失败自动降级双路；需 bge-m3）
+- **记忆管理**：写入 / 更新 / 删除 / 列表，支持 namespace、tags、type 过滤
+- **记忆全共享**：所有记忆/知识与任何客户端、任务、AI 全共享——本地单用户定位，跨 agent、跨任务读到同一份用户记忆；client 仅用于审计归类与元数据；主键服务端生成
+- **全局 + 分项目记忆（v2.1）**：不传 `project` 写入全局库；传 `project` 写入独立子库 `kb_data/projects/<project>/`（物理隔离，删目录即删项目记忆）；带项目检索 = 项目子库 + 全局库加权合并（`KB_PROJECT_BOOST`）
+- **PyPI 安装与依赖瘦身（v2.1）**：`pip install "kb-memory[ingest]"`，摄取/LLM 拆为可选 extras；Linux 可先装 CPU 版 torch，总下载量显著下降
+- **Skill 自动安装与发现（v2.1）**：skill 内置 `bootstrap.py`，**询问用户同意后**才建独立环境安装；`kb where` 告诉你装在哪，`kb autostart enable` 开机自启
+- **空闲卸载（v2.1）**：`KB_IDLE_UNLOAD_SECONDS` 到期自动卸载嵌入模型并释放显存，后台占用趋零
+- **记忆语言与自定义规则（v2.1）**：`KB_MEMORY_LANG=auto` 跟随操作系统语言存储记忆；`~/.kb/rules.md` 自定义"什么时候该存记忆"
+- **存取审计**：每次写/读/改/删/检索/问答记 JSON 到 `logs/agent-audit/<客户端>__<项目>.log`（按 client+project 分文件）；用户可查：REST `GET /api/v1/audit?client=<客户端>[&project=<项目>]` 或 CLI `kb audit --client <客户端>`
+- **知识入库**：本地文档（txt/md/pdf/docx 等）上传或路径导入，网页正文抓取入库
+- **目录监听**：指定目录内新增/删除文件自动入库/清理（`KB_WATCH_DIR`）
+- **CLI 工具**：`kb add/search/stats/ask/eval/forget/dedup`——终端直接完成写入、检索、统计与 RAG 问答
+- **隐私护栏**：敏感 namespace 强制本地回答不出网；`/ask` 智能路由（本地优先，难题可选云端）
+- **断网可用**：模型与数据全部落本地，无网络时存取与检索功能完整
+
+## 快速开始（Windows / Linux / macOS）
+
+**从 PyPI 安装（推荐）：**
+
+```bash
+pip install "kb-memory[ingest]"   # 核心 + 文档/网页摄取；[llm]=云端问答，[full]=全部
+python -m kb serve
+```
+
+**Linux 无显卡机器先装 CPU 版 torch**（PyPI 默认拉 CUDA 轮子，体积大得多）：
+
+```bash
+pip install torch --index-url https://download.pytorch.org/whl/cpu
+pip install "kb-memory[ingest]"
+python -m kb serve
+```
+
+**源码开发安装：**
+
+```powershell
+python -m venv venv
+.\venv\Scripts\Activate.ps1          # Linux/macOS: source venv/bin/activate
+pip install -r requirements.txt
+pip install -e ".[full]"             # 开发/测试需要摄取、监听与 LLM 可选依赖
+python -m kb serve
+```
+
+> **首次安装要下载什么、大概多久**（先看一眼，避免以为卡死了）：
+> - 核心 Python 依赖约 **400–600 MB**（Windows CPU 版 `torch` 约 124 MB 为最大单体；
+>   Linux 先装 CPU 轮子可避免默认 CUDA 轮子带来的数 GB 下载）；
+> - 可选 `[ingest]` 额外约几十 MB（PDF/DOCX/Office 解析、网页提取、目录监听）；
+> - **首次写入/检索时**才加载并下载嵌入模型（懒加载，启动本身很轻），**默认 `bge-small-zh-v1.5` 约 92 MB**（若显式配了 `KB_EMBED_MODEL=BAAI/bge-m3`，则约 2.2 GB）；
+> - 国内网络建议加 pip 镜像（实测阿里云可用、清华可能 403）：
+>   `pip install "kb-memory[ingest]" -i https://mirrors.aliyun.com/pypi/simple/`
+> - 模型下载走 HuggingFace 镜像：`$env:HF_ENDPOINT='https://hf-mirror.com'`（Windows）/ `export HF_ENDPOINT=https://hf-mirror.com`（Linux、macOS）。
+>
+> 重要：用源码 venv 安装时 `kb` 命令只装在虚拟环境内。**每个新终端都要先激活虚拟环境**
+> （Windows `.\venv\Scripts\Activate.ps1`，Linux/macOS `source venv/bin/activate`），
+> 否则会提示 `kb：未找到命令`。改用 `python -m kb <子命令>` 可绕过激活。
+> 不知道自己的 kb 装在哪里、怎么启动：`kb where`，或运行 skill 包里的
+> `python bootstrap.py --status` / `--start-only`。
+
+启动后健康检查：
+
+```powershell
+curl http://127.0.0.1:8000/api/v1/healthz
+```
+
+> 嵌入模型采用懒加载：启动后空闲时几乎不占内存，**首次写入或检索**才加载
+> `BAAI/bge-small-zh-v1.5`（约 92 MB，自动下载并缓存，之后断网可用）；
+> 未配置 LLM 时服务照常启动，`/ask` 返回 503 与配置指引。
+> `/healthz` 的 `embedder` 字段会显示 `unloaded/loading/ok/error`，模型加载失败时
+> 写/查请求返回 **503 `EMBED_MODEL_UNAVAILABLE`**（含模型名与修复建议），不再裸 500。
+>
+> **CPU 即可运行，不占显卡**：全新安装默认拿到 CPU 版 torch（`device=cpu`），
+> 不占用 GPU 显存，无独立显卡的机器也能完整使用。有显卡想加速（大库、批量入库）
+> 可另装 CUDA 版 torch，见 [AGENTS.md](AGENTS.md) 第 7 节；也可显式 `KB_DEVICE=cuda`。
+>
+> **LLM 默认关闭（`KB_LLM_MODE=off`）**：服务启动不会探测/加载/调用任何大模型，
+> 零显存、零成本、纯离线；记忆写入、文档入库、混合检索完整可用。
+> 需要 RAG 问答（`/ask`）时再按下面方式配置本地或云端 LLM。
+
+### 嵌入模型怎么选
+
+默认用轻量模型，是为了让"先试试"的成本足够低；模型不强绑定，可随时按需切换：
+
+| 模型 | 体积 | 维度 | 加载后常驻内存 | 适用 |
+|---|---|---|---|---|
+| **`BAAI/bge-small-zh-v1.5`（默认）** | **约 92 MB** | 512 | 约 100 MB 量级 | 日常偏好/决策/笔记类记忆；CPU 上也快 |
+| `BAAI/bge-m3` | 约 2.2 GB | 1024 | 约 2 GB | 长文档、跨语言、语义更微妙的内容，需要更强语义召回 |
+
+> 本项目在自带 50 题中文评测集上的实测（`kb eval`）：轻量模型 Recall@1 **0.760** / MRR **0.847** /
+> 平均延迟 **28.6 ms**；bge-m3 为 0.740 / 0.841 / 105.6 ms。小样本下二者基本持平，轻量模型更快更省。
+> 但 **bge-m3 在长文档与跨语言场景预期更强**，两个模型定位不同，按自己数据选。
+
+**换模型前必读**：向量维度不同（512 vs 1024），ChromaDB 集合维度由首次写入的向量决定，
+**换了模型后旧数据无法直接检索**。所以：
+
+- **新库**（还没写过数据）：直接设 `KB_EMBED_MODEL=BAAI/bge-m3` 即可；
+- **已有数据的库**：kb 会自动**沿用建库时使用的模型**（记录在 `kb_data/runtime.json`），
+  不会因默认值变化而破坏你的数据，启动日志会记录实际沿用的模型；
+  若能识别的配置与库不一致（你显式改了 `KB_EMBED_MODEL`），还会给出 WARNING 说明如何迁移。
+  确要换模型，请先备份，再清空 `kb_data/chroma` 后重新入库。
+
+### 配置 LLM（可选，`/ask` 问答需要）
+
+`KB_LLM_MODE` 默认 `off`（不加载不调用）；四个档位：
+
+| 档位 | 行为 |
+|---|---|
+| `off`（默认） | 完全不加载/不调用 LLM；记忆存取与检索完整可用 |
+| `local` | 仅本地 Ollama（完全离线，隐私零出网） |
+| `auto` | **本地优先，云端降级**：本地 Ollama 可用走本地，无本地但有云端 Key 走云端 |
+| `cloud` | 全部走云端（本地仅做压缩与隐私隔离） |
+
+**本地 LLM（Ollama）：**
+
+```powershell
+# 1. 安装并启动 Ollama（Windows 从开始菜单/托盘启动，不要从 AI 沙箱终端拉起）
+# 2. 拉取一个适合你电脑的模型（按显存/内存选择，如 qwen3:4b 约 3.2GB、
+#    qwen3:1.7b 约 1.8GB；国内可用魔搭加速，拉完 ollama cp 改成短名）
+ollama pull <你的模型名>
+# 3. 在 .env 配置后重启服务：
+#    KB_LLM_MODE=local            # 仅本地
+#    KB_LLM_MODEL=<你的模型名>    # 以 `ollama list` 输出的名称为准
+#    KB_OLLAMA_BASE_URL=http://localhost:11434
+```
+
+**云端 LLM（任意 OpenAI 兼容服务商，不绑定 DeepSeek）：**
+DeepSeek / OpenAI / 通义千问 / 硅基流动 / Moonshot 等均可，通用三键：
+
+```ini
+# .env
+KB_LLM_MODE=auto                 # 本地优先、云端降级；或 cloud 全云端
+KB_LLM_API_KEY=sk-xxx            # 服务商 API Key
+KB_LLM_BASE_URL=https://api.deepseek.com   # 换成你所用服务商的 OpenAI 兼容端点
+KB_LLM_CLOUD_MODEL=deepseek-v4-flash       # 云端模型名
+```
+
+验证：`GET /api/v1/healthz` 的 `llm` 字段——`local`/`cloud` 表示 LLM 已就绪，`disabled` 表示未启用。
+
+### 常见问题：嵌入模型下载失败
+
+- **中国大陆直连 `huggingface.co` 会超时**。设置 HF 镜像后重启即可：
+  ```bash
+  export HF_ENDPOINT=https://hf-mirror.com   # 或写入 ~/.bashrc 永久生效
+  python -m kb serve
+  ```
+  模型会自动从镜像下载并缓存到 `~/.cache/huggingface/hub/`，之后断网也能离线加载。
+- **模型已在本机缓存，但无外网**：`kb` 采用离线优先（先命中本地缓存，失败才联网），
+  只要缓存目录完整即可完全离线运行。
+
+## 让 Agent 接入 kb（客户端无关）
+
+把 kb 的接入规约交给 AI 客户端（TraeWork / Claude Code / Cursor / 自建 Agent），
+让它们知道怎么读写记忆、按什么身份规约、怎么查审计。**两种方式，任选其一**：
+
+1. **skill（推荐，能自动触发）——可选的独立步骤**：仓库内
+   [`skills/kb-memory/SKILL.md`](skills/kb-memory/SKILL.md)
+   是客户端无关的 Anthropic 开格式 skill。把它安装到你所用客户端的用户级 skills 目录后，
+   该客户端的任何项目会话都会在读写记忆/RAG 问答/审计查询时**自动识别并触发**。
+   安装 = 把 `skills/kb-memory` 目录复制过去即可（有脚本，也可手动复制，无需任何依赖）；
+   更新（重新覆盖）、卸载、以及装好后的使用说明见
+   [`scripts/README.md`](scripts/README.md)。
+
+   > **不装也不影响 kb 服务**：skill 只是给 AI 客户端的「提示词包装」，与服务的安装、
+   > 启动无关——跳过这一步，服务照常运行，你随时可用方法 2 的纯文本提示词接入；
+   > skill 安装是**一次性、按需、独立执行**的，不会随 `kb serve` 自动触发，
+   > 也不会写入你的任何客户端目录以外的文件。
+2. **纯文本提示词（兜底，任何客户端通用）**：整段复制
+   [`docs/AGENT_PROMPT.md`](docs/AGENT_PROMPT.md) 粘贴给 Agent 即可，不依赖 skill 机制。
+
+> **`.trae-cn/skills` / `.claude/skills` / `.cursor/skills` 是任何客户端都认的「标准」吗？——不是。**
+> 这些只是各家客户端各自的**用户级约定目录**：`SKILL.md` 本身是统一的 Anthropic 开格式，
+> 但「装到哪个目录、能否自动触发」由各客户端自行决定，支持程度不一：
+
+| 客户端 | 用户级 skills 目录 | 自动加载 |
+|---|---|---|
+| TraeWork | `~/.trae-cn/skills/` | 自动发现 |
+| Claude Code | `~/.claude/skills/` | 高版本支持 |
+| Cursor | `~/.cursor/skills/` | 逐步跟进 |
+| 其他 / 自建 Agent | 无统一约定 | 需手动加载或不支持 |
+
+> 不存在「所有客户端都遵循」的统一目录；你的客户端若不支持 skill，**永远有方法 2 兜底**
+> （粘贴 `AGENT_PROMPT.md`，纯文本任何客户端可用）。安装方法、各客户端目录差异与加载机制、
+> 相互引用关系：详细见 [`scripts/README.md`](scripts/README.md)（此处不重复）。
+
+### 打包为独立 skill（Agensi / marketplace）
+
+`skills/kb-memory/` 是 **kb 的独立交付单元**，只依赖 kb 服务本身：
+包含标准 front matter 的 `SKILL.md`（包根）、MCP 配置模板 `mcp.json`、配置模板
+`env.example`、独立 README 与 LICENSE，覆盖「安装 → 启动 → 配置 → 验证 → 排障」全流程。
+
+```powershell
+python scripts/build_skill_package.py        # 产物：dist/kb-memory-skill-<版本>.zip
+```
+
+zip 内 `SKILL.md` 位于**包根目录**，不含测试、运行数据或任何密钥；
+构建脚本会校验 front matter、自包含说明与敏感信息。可直接提交到 Agensi 这类
+skill marketplace，用户拿到包后按 `SKILL.md` 即可把本机 kb 接入自己的 Agent。
+
+## MCP 挂载
+
+MCP 端点（streamable HTTP）：`http://127.0.0.1:8000/mcp/`
+
+**Claude Code**：本仓库已内置项目级 `.mcp.json`，在本目录启动 Claude Code 即自动挂载；
+也可全局添加：
+
+```powershell
+claude mcp add --transport http kb http://127.0.0.1:8000/mcp/
+```
+
+**Cursor / TraeWork 及其他支持 MCP 的客户端**：在 MCP 配置中加入以下 JSON
+（Cursor 放 `~/.cursor/mcp.json` 或项目 `.cursor/mcp.json`；TraeWork 在设置中添加 MCP 服务器）：
+
+```json
+{
+  "mcpServers": {
+    "kb": {
+      "type": "http",
+      "url": "http://127.0.0.1:8000/mcp/"
+    }
+  }
+}
+```
+
+挂载后可用的 MCP 工具：`write_memory` / `search_memory` / `read_memory` /
+`update_memory` / `delete_memory` / `add_document` / `add_webpage` / `ask_kb`。
+
+> 启用 `KB_API_KEY` 鉴权后，MCP 客户端需在连接配置加 `headers`（`Authorization: Bearer <key>`）；
+> 仓库内 `.mcp.json` 模板不含真实 key（JSON 不支持注释），配法见 [USER_GUIDE §5.2](docs/USER_GUIDE.md#52-客户端适配n20启用鉴权后各客户端如何带-key)。
+
+## REST 端点速查
+
+| 方法 | 路径 | 说明 |
+|---|---|---|
+| POST | `/api/v1/memories` | 写入记忆 `{content, tags?, source?, namespace?}` |
+| GET | `/api/v1/memories` | 记忆列表，支持 `type/tag/source/q/limit/offset` 过滤 |
+| GET | `/api/v1/memories/{id}` | 读取单条记忆 |
+| PATCH | `/api/v1/memories/{id}` | 更新内容或标签 |
+| DELETE | `/api/v1/memories/{id}` | 删除单条记忆 |
+| POST | `/api/v1/search` | 混合检索 `{query, top_k?, mode?, type?, tag?}`，mode: `hybrid/vector/keyword` |
+| POST | `/api/v1/documents` | 文档入库：multipart `file` 字段或 JSON `{"path": "本地路径"}` |
+| GET | `/api/v1/documents` | 已入库文档列表（按 source 聚合） |
+| DELETE | `/api/v1/documents/{source}` | 按 source 删除该文档全部记录（本地文件名等简单 source） |
+| DELETE | `/api/v1/documents?source=<urlencoded>` | 按查询参数删除（**URL 等含斜杠的 source 必须用这个**） |
+| POST | `/api/v1/ingest/web` | 网页入库 `{url}`，抓取正文切分入库；同 URL 重复入库自动替换旧记录 |
+| POST | `/api/v1/ask` | RAG 问答 `{question}`；未配置 LLM 返回 503 |
+| GET | `/api/v1/healthz` | 健康检查与服务统计（含 `embedder` 状态） |
+| GET | `/api/v1/governance/stats` | 记忆治理统计：`total_count`/`avg_access_count`/`stale_90d_count`（只读） |
+| GET | `/api/v1/governance/config` | 治理配置：衰减+新鲜度开关与参数（只读） |
+| POST | `/api/v1/memories` → **409** | 启用语义去重（`KB_DEDUP_ENABLED=true`）后写入命中重复返回 409：`{"error":"DUPLICATE","duplicate_of":"<已有记录id>","similarity":<相似度>}`（不写入） |
+
+> 记忆治理（去重/衰减/新鲜度）均默认关闭、零行为变化；用法见 [USER_GUIDE §3.5](docs/USER_GUIDE.md#35-记忆治理a3语义去重--衰减--新鲜度)。
+
+示例（Windows PowerShell 可直接复制；中文必须按 UTF-8 字节发送）：
+
+```powershell
+function Invoke-KbJson($Uri, $Body) {
+  Invoke-RestMethod -Uri $Uri -Method Post -ContentType 'application/json; charset=utf-8' `
+    -Body ([System.Text.Encoding]::UTF8.GetBytes(($Body | ConvertTo-Json -Depth 5)))
+}
+
+# 写入一条记忆
+Invoke-KbJson http://127.0.0.1:8000/api/v1/memories @{ content = "用户偏好深色主题"; tags = @("偏好") }
+
+# 混合检索
+Invoke-KbJson http://127.0.0.1:8000/api/v1/search @{ query = "用户界面偏好"; top_k = 5 }
+
+# 按 URL 删除网页文档（source 需 URL 编码；PowerShell 用 [uri]::EscapeDataString）
+$src = [uri]::EscapeDataString("https://example.com/page")
+Invoke-RestMethod -Uri "http://127.0.0.1:8000/api/v1/documents?source=$src" -Method Delete
+```
+
+Linux / macOS 用 `curl`（`curl` 在这里是原生命令，不是 PowerShell 别名）：
+
+```bash
+curl -X POST http://127.0.0.1:8000/api/v1/memories \
+  -H "Content-Type: application/json" \
+  -d '{"content": "用户偏好深色主题", "tags": ["偏好"]}'
+```
+
+## 配置项简表
+
+全部配置以 `KB_` 前缀的环境变量或 `.env` 文件提供；完整键名见 [`.env.example`](.env.example)
+（复制为 `.env` 后填写，`.env` 已被 gitignore，真实密钥只放本机，严禁入库）。
+
+| 配置项 | 默认值 | 说明 |
+|---|---|---|
+| `KB_LLM_MODE` | `off` | LLM 模式：`off`（默认，不加载/不调用 LLM，零显存零成本）/ `local`（仅本地 Ollama）/ `auto`（本地优先，云端降级）/ `cloud` |
+| `KB_DEVICE` | 空 | 嵌入模型设备：空=自动检测，可显式设 `cpu` / `cuda` |
+| `KB_WATCH_DIR` | `data` | serve 模式监听目录，文件变动自动入库；空串或 `.` = 不启动 |
+| `KB_DATA_DIR` | `kb_data` | 运行数据根目录（ChromaDB、运行时状态等） |
+| `KB_API_HOST` / `KB_API_PORT` | `127.0.0.1` / `8000` | REST 与 MCP 监听地址 |
+| `KB_EMBED_MODEL` | `BAAI/bge-small-zh-v1.5` | 嵌入模型（轻量默认，约 92 MB）；可换 `BAAI/bge-m3`（约 2.2 GB，更强），见「嵌入模型怎么选」 |
+| `KB_LLM_MODEL` | 空 | 本地 Ollama 模型名（默认空=不配；配 `KB_LLM_MODE=local/auto` 时须按自己电脑选模型，以 `ollama list` 为准） |
+| `KB_OLLAMA_BASE_URL` | `http://localhost:11434` | Ollama 端点 |
+| `KB_LLM_API_KEY` / `KB_LLM_BASE_URL` / `KB_LLM_CLOUD_MODEL` | 空 | 云端 LLM（可选）：**任意 OpenAI 兼容服务商**（DeepSeek / OpenAI / 通义 / 硅基流动等），仅填在本机 `.env` |
+| `KB_CHUNK_SIZE` / `KB_CHUNK_OVERLAP` | `500` / `100` | 文档切分参数 |
+| `KB_SENSITIVE_NAMESPACES` | 空 | 逗号分隔的敏感 namespace，命中强制本地回答不出网 |
+| `KB_API_KEY` | 空 | 空=不鉴权（本地回环零摩擦）；非空=启用 Bearer/X-API-Key 鉴权 |
+| `KB_RERANK_ENABLED` / `KB_RERANK_MODEL` / `KB_RERANK_TOP_N` | `false` / `BAAI/bge-reranker-v2-m3` / `20` | 检索精排（A3.5）：融合候选送 CrossEncoder 重排，默认关 |
+| `KB_SPARSE_ENABLED` | `false` | 稀疏第三路（A3.5）：BGE-M3 稀疏向量 + 倒排索引参与归一化加权融合，默认关 |
+
+## CLI 速查（无需启动服务）
+
+```powershell
+python -m kb add "记忆内容" --tags 偏好 --client TraeWork                       # 写入（project 缺省自动取当前目录名）
+python -m kb search "查询词" --client TraeWork                                 # 混合检索（v2.1：不传 project 查全局；传 project 合并子库）
+python -m kb stats                            # 统计：类型分布 / 访问热度 / 陈旧分布
+python -m kb ask "问题" --client TraeWork      # 终端 RAG 问答（LLM 不可用时输出检索命中）
+python -m kb audit --client TraeWork --days 7   # 查某客户端/项目存过/读过什么
+python -m kb eval --file tests/eval_zh_50.jsonl   # 检索质量评测（Recall@1/@5 + MRR）
+python -m kb forget --stale --days 90         # 清理超期未命中记忆
+python -m kb dedup --threshold 0.92           # 语义去重
+```
+
+> `kb ask` 直连本地服务逻辑（不经 HTTP）；建议 `serve` 停止时使用，避免双进程写库竞争。
+
+## 目录结构
+
+```
+kb/            kb 服务源码（config / models / embedder / storage / bm25 / retriever /
+               service / llm / ingest / watcher / api / mcp / cli + reranker / sparse / eval）
+tests/         kb 验收测试（417 项通过 / 1 跳过，含 eval_zh_50.jsonl 检索评测数据集）
+docs/          核心文档、当前设计（历史 spec/plan 已归档本地，不上传）
+kb_data/       kb 运行数据（gitignore）
+_archive/      旧学习项目 + 历史 spec/plan 归档（仅本地保留，禁止参考）
+```
+
+## 更多文档
+
+- **用户使用手册（人类用户入口）**：[docs/USER_GUIDE.md](docs/USER_GUIDE.md)
+- **AI 接力文档（AI 助手入口）**：[PROJECT.md](PROJECT.md)（项目状态 / 进度看板 / 接手指南）
+- 设计文档（需求、架构、API、里程碑）：`docs/superpowers/specs/2026-08-23-kb-memory-service-design.md`
+- **v2.1 设计（当前）**：`docs/superpowers/specs/2026-10-20-kb-v2.1-design.md`
+- 开发者维护手册：`docs/DEV_HANDBOOK.md`
+- AI 协作规范：`AGENTS.md`
+- 历史 spec / 节点计划已归档到本地 `_archive/docs/`（不上传，不再作为开发依据）