{"title":"记忆打标签策略","slug":"memory-tagging-strategy","category":"llm-cookbook","summary":"如何为记忆打标签以实现高效搜索与过滤 — 可扩展的标签体系。","audience":["llm"],"tags":["cookbook","tagging","memory","strategy"],"difficulty":"intermediate","updated":"2026-06-27","word_count":168,"read_minutes":1,"lang":"zh","translated":true,"requested_lang":"zh","content_markdown":"\n# 记忆打标签策略\n\n标签是可扩展记忆检索的秘诀。本指南展示如何为记忆打标签，让正确的记忆在正确的时间被找到。\n\n## 为什么标签重要\n\n没有标签，你只有扁平的全文搜索。有了标签，你就有结构化导航：\n\n```bash\n# 无标签：搜索一切\nGET /memory/search?q=docker\n\n# 有标签：按项目 + 技术过滤\nGET /memory/search?q=deployment&tag=synapse&tag=docker\n```\n\n标签带来：\n\n- **快速过滤** — `GET /memory/by-tag?tag=production`\n- **范围搜索** — `?q=auth&tag=project-x`\n- **分组** — 找出某项目的所有 “mistake” 记忆\n- **交叉引用** — 共享标签的记忆相互关联\n\n## 标签体系\n\n### 项目标签\n\n用项目名作为标签：\n\n```\nsynapse, synapse-mcp, synapse-chat, synapse-sdk\n```\n\n### 技术标签\n\n用技术名：\n\n```\ndocker, kubernetes, postgres, fastify, react, typescript\n```\n\n### 主题标签\n\n用主题分类：\n\n```\ndeployment, ci-cd, auth, database, frontend, backend, security\n```\n\n### 状态标签\n\n用状态指示：\n\n```\nactive, completed, blocked, deprecated\n```\n\n### 类型标签\n\n用类型指示：\n\n```\ndecision, mistake, pattern, reference, todo\n```\n\n## 打标签规则\n\n### 规则 1：每条记忆 2-5 个标签\n\n太少 = 可发现性差。太多 = 噪声。\n\n```json\n// 好：3 个相关标签\n{ \"tags\": [\"synapse\", \"deployment\", \"docker\"] }\n\n// 坏：1 个标签（太窄）\n{ \"tags\": [\"synapse\"] }\n\n// 坏：10 个标签（噪声）\n{ \"tags\": [\"synapse\", \"deployment\", \"docker\", \"vps1\", \"2026\", \"june\", \"ssh\", \"git\", \"main\", \"production\"] }\n```\n\n### 规则 2：小写、连字符\n\n```\n✅ ci-cd, api-key, mind-key\n❌ CI-CD, APIKey, MindKey\n```\n\n### 规则 3：使用一致的词汇\n\n建立一套标签词汇并坚持使用：\n\n```\n# 项目词汇\nsynapse, synapse-mcp, synapse-chat\n\n# 不要写成：synapse_project, synapseProject, SYNAPSE\n```\n\n### 规则 4：按搜索意图打标签\n\n问自己：“我以后会怎么搜这条记忆？”\n\n```json\n// 存储部署决策\n{\n \"content\": \"Decided to use Docker Swarm for Synapse deployment\",\n \"tags\": [\"synapse\", \"deployment\", \"docker\", \"swarm\", \"decision\"]\n}\n\n// 你可能会搜索：?q=docker+swarm 或 ?tag=deployment\n```\n\n## 模式\n\n### 模式 1：项目 + 主题\n\n```json\n{ \"tags\": [\"synapse\", \"deployment\"] }\n{ \"tags\": [\"synapse\", \"auth\"] }\n{ \"tags\": [\"synapse-mcp\", \"tools\"] }\n```\n\n搜索：`?tag=synapse`（所有 Synapse 项目记忆）\n搜索：`?tag=synapse&q=deployment`（Synapse 的部署记忆）\n\n### 模式 2：类型 + 领域\n\n```json\n{ \"tags\": [\"mistake\", \"deployment\"] }\n{ \"tags\": [\"decision\", \"database\"] }\n{ \"tags\": [\"pattern\", \"auth\"] }\n```\n\n搜索：`?tag=mistake`（所有错误）\n搜索：`?tag=mistake&q=deployment`（部署错误）\n\n### 模式 3：层级式\n\n用于项目内的子项目：\n\n```json\n{ \"tags\": [\"synapse\", \"synapse-docs\", \"markdown\"] }\n{ \"tags\": [\"synapse\", \"synapse-mcp\", \"mcp\"] }\n{ \"tags\": [\"synapse\", \"synapse-admin\", \"ui\"] }\n```\n\n搜索：`?tag=synapse`（所有 Synapse）\n搜索：`?tag=synapse-docs`（仅 docs 子项目）\n\n### 模式 4：状态跟踪\n\n```json\n// 活跃项目\n{ \"tags\": [\"synapse\", \"active\"], \"priority\": \"high\" }\n\n// 已完成项目\n{ \"tags\": [\"synapse-v1\", \"completed\"], \"priority\": \"low\" }\n\n// 阻塞中\n{ \"tags\": [\"synapse-v2\", \"blocked\"], \"priority\": \"high\" }\n```\n\n## 常见用例\n\n### 找出某项目的所有决策\n\n```bash\ncurl -H \"Authorization: Bearer $KEY\" \\\n \".../memory/search?q=decision&tag=synapse\"\n```\n\n### 找出某领域的所有错误\n\n```bash\ncurl -H \"Authorization: Bearer $KEY\" \\\n \".../memory/search?q=mistake&tag=deployment\"\n```\n\n### 找活跃工作\n\n```bash\ncurl -H \"Authorization: Bearer $KEY\" \\\n \".../memory/by-tag?tag=active\"\n```\n\n### 找关于某技术的记忆\n\n```bash\ncurl -H \"Authorization: Bearer $KEY\" \\\n \".../memory/search?q=postgres+performance&tag=database\"\n```\n\n## 标签维护\n\n### 定期审查\n\n```python\n# 找出使用少的标签（清理候选）\ntags = requests.get(f\"{URL}/memory/tags\",\n headers={\"Authorization\": f\"Bearer {KEY}\"}).json()\n\nfor tag, count in tags.items():\n if count < 2:\n print(f\"Rare tag: {tag} ({count} memories)\")\n```\n\n### 合并标签\n\n如果有不一致的标签（`docker` 和 `Docker`），合并它们：\n\n```python\n# 找出所有带 \"Docker\" 标签的记忆\nmems = requests.get(f\"{URL}/memory/by-tag?tag=Docker\",\n headers={\"Authorization\": f\"Bearer {KEY}\"}).json()\n\n# 把每条更新为 \"docker\"\nfor mem in mems[\"results\"]:\n tags = [t.lower() for t in mem[\"tags\"]]\n update_memory(mem[\"id\"], tags=list(set(tags)))\n```\n\n## 最佳实践\n\n> [!TIP]\n> - **尽早建立词汇表** — 从第 1 天起就用一致的标签\n> - **按搜索意图打标签** — 你以后会怎么找它？\n> - **2-5 个标签是最佳点** — 太少或太多都不好\n> - **小写 + 连字符** — `ci-cd` 而非 `CI/CD`\n> - **定期审查** — 合并重复、移除未用\n\n## 下一步\n\n- [记忆最佳实践](/docs/guides/memory-best-practices)\n- [FTS5 搜索](/docs/concepts/fts5-search)\n- [任务驱动工作流](/docs/llm-cookbook/task-driven-workflow)\n","content_html":"

记忆打标签策略

\n

标签是可扩展记忆检索的秘诀。本指南展示如何为记忆打标签，让正确的记忆在正确的时间被找到。

\n

为什么标签重要

\n

没有标签，你只有扁平的全文搜索。有了标签，你就有结构化导航：

\n
# 无标签：搜索一切\nGET /memory/search?q=docker\n\n# 有标签：按项目 + 技术过滤\nGET /memory/search?q=deployment&tag=synapse&tag=docker

标签带来：

\n\n

标签体系

\n

项目标签

\n

用项目名作为标签：

\n
synapse, synapse-mcp, synapse-chat, synapse-sdk

技术标签

\n

用技术名：

\n
docker, kubernetes, postgres, fastify, react, typescript

主题标签

\n

用主题分类：

\n
deployment, ci-cd, auth, database, frontend, backend, security

状态标签

\n

用状态指示：

\n
active, completed, blocked, deprecated

类型标签

\n

用类型指示：

\n
decision, mistake, pattern, reference, todo

打标签规则

\n

规则 1：每条记忆 2-5 个标签

\n

太少 = 可发现性差。太多 = 噪声。

\n
// 好：3 个相关标签\n{ "tags": ["synapse", "deployment", "docker"] }\n\n// 坏：1 个标签（太窄）\n{ "tags": ["synapse"] }\n\n// 坏：10 个标签（噪声）\n{ "tags": ["synapse", "deployment", "docker", "vps1", "2026", "june", "ssh", "git", "main", "production"] }

规则 2：小写、连字符

\n
✅ ci-cd, api-key, mind-key\n❌ CI-CD, APIKey, MindKey

规则 3：使用一致的词汇

\n

建立一套标签词汇并坚持使用：

\n
# 项目词汇\nsynapse, synapse-mcp, synapse-chat\n\n# 不要写成：synapse_project, synapseProject, SYNAPSE

规则 4：按搜索意图打标签

\n

问自己：“我以后会怎么搜这条记忆？”

\n
// 存储部署决策\n{\n  "content": "Decided to use Docker Swarm for Synapse deployment",\n  "tags": ["synapse", "deployment", "docker", "swarm", "decision"]\n}\n\n// 你可能会搜索：?q=docker+swarm 或 ?tag=deployment

模式

\n

模式 1：项目 + 主题

\n
{ "tags": ["synapse", "deployment"] }\n{ "tags": ["synapse", "auth"] }\n{ "tags": ["synapse-mcp", "tools"] }

搜索：?tag=synapse（所有 Synapse 项目记忆）\n搜索：?tag=synapse&q=deployment（Synapse 的部署记忆）

\n

模式 2：类型 + 领域

\n
{ "tags": ["mistake", "deployment"] }\n{ "tags": ["decision", "database"] }\n{ "tags": ["pattern", "auth"] }

搜索：?tag=mistake（所有错误）\n搜索：?tag=mistake&q=deployment（部署错误）

\n

模式 3：层级式

\n

用于项目内的子项目：

\n
{ "tags": ["synapse", "synapse-docs", "markdown"] }\n{ "tags": ["synapse", "synapse-mcp", "mcp"] }\n{ "tags": ["synapse", "synapse-admin", "ui"] }

搜索：?tag=synapse（所有 Synapse）\n搜索：?tag=synapse-docs（仅 docs 子项目）

\n

模式 4：状态跟踪

\n
// 活跃项目\n{ "tags": ["synapse", "active"], "priority": "high" }\n\n// 已完成项目\n{ "tags": ["synapse-v1", "completed"], "priority": "low" }\n\n// 阻塞中\n{ "tags": ["synapse-v2", "blocked"], "priority": "high" }

常见用例

\n

找出某项目的所有决策

\n
curl -H "Authorization: Bearer $KEY" \\\n     ".../memory/search?q=decision&tag=synapse"

找出某领域的所有错误

\n
curl -H "Authorization: Bearer $KEY" \\\n     ".../memory/search?q=mistake&tag=deployment"

找活跃工作

\n
curl -H "Authorization: Bearer $KEY" \\\n     ".../memory/by-tag?tag=active"

找关于某技术的记忆

\n
curl -H "Authorization: Bearer $KEY" \\\n     ".../memory/search?q=postgres+performance&tag=database"

标签维护

\n

定期审查

\n
# 找出使用少的标签（清理候选）\ntags = requests.get(f"{URL}/memory/tags",\n    headers={"Authorization": f"Bearer {KEY}"}).json()\n\nfor tag, count in tags.items():\n    if count < 2:\n        print(f"Rare tag: {tag} ({count} memories)")

合并标签

\n

如果有不一致的标签（docker 和 Docker），合并它们：

\n
# 找出所有带 "Docker" 标签的记忆\nmems = requests.get(f"{URL}/memory/by-tag?tag=Docker",\n    headers={"Authorization": f"Bearer {KEY}"}).json()\n\n# 把每条更新为 "docker"\nfor mem in mems["results"]:\n    tags = [t.lower() for t in mem["tags"]]\n    update_memory(mem["id"], tags=list(set(tags)))

最佳实践

\n

下一步

\n\n","urls":{"html":"/docs/llm-cookbook/memory-tagging-strategy","text":"/docs/llm-cookbook/memory-tagging-strategy?format=text","json":"/docs/llm-cookbook/memory-tagging-strategy?format=json","llm":"/docs/llm-cookbook/memory-tagging-strategy?format=llm"},"translations_available":["en","zh","hi","es","fr","ar","pt","ru","ja","de","it","ko","nl","pl","tr","sv","vi","th","id","uk"]}