蓝汐 · 潮汐认知图谱
Supports read-only import of external Markdown and Obsidian vault content into the local memory graph via the cg_import_vault tool.
Provides optional AI enhancement through OpenAI-compatible chat completions and embeddings APIs, enabling semantic vector search and AI-assisted condensation of memory nodes.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@蓝汐 · 潮汐认知图谱 回忆一下我们之前定下的数据库选型,并列出相关记忆节点"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
蓝汐 · 潮汐认知图谱(Lanxi Memory)
简体中文 | English
Local-first, zero-dependency cognitive graph memory plugin for Antigravity — 5-layer Markdown vault + CJK FTS5 + Tidal forgetting curves + Optional AI semantic enhancement + Tidal Starfield UI.
蓝汐(Lanxi) 是一款专为 Google Antigravity 打造的纯本地长期记忆与认知图谱插件。它把「人类可读的纯 Markdown 真源文档」与「幂律/指数潮汐遗忘曲线」融合在一起:重要约定永久锚定不衰减,日常情景随时间退潮、被频繁检索则涨潮强化,绝不静默丢弃任何记忆文件。
✨ 核心特性
零外部依赖 · 开箱即用:仅依赖 Node.js(
>= 22.0.0)内置的node:sqlite(WAL + FTS5)与原生fetch,无需安装 Python、Docker 或外置向量库。五层认知结构 + 硬保护机制:
anchor(永久锚点)、self(自我认知)、goals(长期目标):定海锚点,永久保留(retention = 1.0)。structural(架构规律)、knowledge(通用知识):importance >= 0.7自动开启硬保护,归档或降级需显式确认override: true。episodic(情景事件):遵循潮汐遗忘曲线自然衰减,召回或强化后自动提升保留度。unresolved(待解决问题):持续追踪未决疑点。
中英混合高精度检索:自研 CJK 2-gram + ASCII 分词器写入 SQLite FTS5,毫秒级精准召回中文长句。
智能生命周期钩子(防噪音 & 防泄露):
SessionStart:注入身份锚点与活跃目标。PreInvocation:过滤常见口语填充词,避免短闲聊误召回无关历史,并支持每 10 轮长会话归档提醒。Stop:自动脱敏常见凭据与私钥后沉淀情景摘要,从不阻塞对话结束。
原创「潮汐星海」工作台(Sidecar + 浏览器全屏双形态):
同心轨道潮汐星图 Canvas(支持拖拽、缩放、悬停查看与点击建边)、潮汐水位分布条、记忆流、锚点目标页与维护台。
可选 AI 增强(默认关闭,完全自愿):
接入任意 OpenAI 兼容接口后,额外解锁 混合语义向量检索(FTS5 + Cosine) 与 AI 智能凝结(自动提炼多条零散情景为高密度知识节点)。不配 API Key 照样 100% 离线完整运行。
Related MCP server: openmemkit
🚀 快速安装(3 步)
前置要求:Node.js >= 22.0.0(内置
node:sqlite)。
克隆仓库到 Antigravity 全局插件目录:
# macOS / Linux git clone https://github.com/A16Nights/lanxi-memory.git ~/.gemini/config/plugins/lanxi-memory # Windows PowerShell git clone https://github.com/A16Nights/lanxi-memory.git "$env:USERPROFILE\.gemini\config\plugins\lanxi-memory"重启 Antigravity(彻底退出托盘后重新打开)。
在客户端 Customizations 页面确认
蓝汐 · 潮汐认知图谱已启用,或在对话中直接点击[蓝汐 · 潮汐星海](sidecar://lanxi-memory/starfield/)打开工作台。
🧰 MCP 工具一览(12 个)
工具名 | 作用 |
| 写入记忆节点(自动去重与 |
| 潮汐加权召回(启用 Embedding 时自动执行 FTS5 + 语义混合排序) |
| 按节点 ID 读取全文、保留度、出入边与磁盘路径 |
| 更新节点内容、层级或重要度(带硬保护拦截) |
| 显式归档到 |
| 在节点之间建立有向语义边(同步写入 frontmatter) |
| 获取全局或局部子图数据 |
| 列出或创建/更新长期目标( |
| 记录元认知反思或未决问题( |
| 维护队列、晋升、强化、手动/AI 凝结、向量补全、索引重建 |
| 获取运行概览、潮汐水位分布与脱敏配置 |
| 严格只读导入外部 Markdown / Obsidian Vault / DSH mdcg |
✨ 可选 AI 增强配置
所有 AI 能力均默认关闭。你可以直接在工作台的 「⚙ 潮汐维护 → 可选 AI 增强接口」 中填写,或通过环境变量注入:
LANXI_LLM_API_KEY:用于 AI 智能凝结(OpenAI 兼容/chat/completions)LANXI_EMBEDDING_API_KEY:用于语义向量混合检索(OpenAI 兼容/embeddings)
密钥仅保存在本机
~/.gemini/lanxi-memory/config.json,接口和界面绝不会回显明文。
🔒 隐私与数据安全
100% 本地运行:真源 Markdown 存储在
~/.gemini/lanxi-memory/vault/,检索索引存储在~/.gemini/lanxi-memory/index.sqlite。除非你主动填写并开启可选 AI 接口,否则插件不发起任何外网请求。
自检命令:
npm test(使用隔离临时目录与模拟服务端进行全链路测试)。
💡 灵感致谢
本项目为基于 Antigravity 插件协议独立设计与编写的原生实现,设计理念受以下优秀开源项目启发,特此致谢:
dsh-memory(灵枢) —— 五层 Markdown 认知图谱、硬保护防失忆与确定性规则理念
CortexGraph —— 类人幂律遗忘曲线与访问强化机制
MIT License © 2026 A16Nights
Available Tools
12 toolscg_forgetA
显式遗忘/归档记忆节点(绝不静默物理删除,仅移入 vault/archive/;硬保护节点需传 override: true)。
Inputs: id (required), reason, override.
cg_goalB
列出或创建/更新长期目标与待办(存储在 goals 层,默认硬保护)。
Inputs: id, op, title, status, +3 more.
cg_graphB
获取记忆关系子图(以某节点为中心或按层级返回关联度最高的节点和边)。
Inputs: limit, layers, centerId.
cg_import_vaultB
以严格只读方式从外部 Markdown 目录(兼容标准 Frontmatter、Obsidian Vault 与 DSH mdcg)导入历史笔记到蓝汐认知图谱。
Inputs: sourceDir (required).
cg_linkB
在两个记忆节点之间建立有向语义关系(同时写入 SQLite 边表与源文件 frontmatter)。
Inputs: sourceId (required), targetId (required), weight, relation.
cg_maintainB
潮汐维护:查看维护队列(queue),晋升情景(promote),强化访问(touch),手动或 AI 智能凝结(consolidate / ai_consolidate),补全向量(backfill_embeddings),或重建索引(rebuild)。
Inputs: id, op, limit, title, +5 more.
cg_readA
按节点 ID 读取完整记忆详情、潮汐保留度、出入边关系与磁盘文件路径。
Inputs: id (required).
cg_recallB
按关键词、分层与潮汐加权检索记忆(若已配置 Embedding API 则自动执行 FTS5 + 向量语义混合召回),并强化命中节点。
Inputs: query (required), limit, layers, includeExpired.
cg_reflectB
记录元认知反思或待解决问题(写入 self 或 unresolved 层,并关联 derived_from 依据节点)。
Inputs: content (required), tags, layer, title, +2 more.
cg_statsA
获取蓝汐(Lanxi)认知图谱运行概览、潮汐水位分布、各层节点统计与配置(密钥已脱敏)。
Inputs: no named parameters.
cg_updateB
更新记忆节点内容、层级、标签或重要度(若节点受硬保护,降低重要度或解除保护需传 override: true)。
Inputs: id (required), tags, layer, title, +4 more.
cg_writeB
写入长期记忆节点到 Markdown 认知图(自动校验与去重;importance >= 0.7 或 anchor/self 层自动开启硬保护)。
Inputs: content (required), tags, layer, title, +4 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 12 tools
Most tools target clearly distinct actions on the memory graph: CRUD (read/write/update/forget), relationship linking/graph traversal, recall, goals, reflection, maintenance, stats, and import. The main overlap is that cg_write, cg_goal, and cg_reflect all create nodes, but their target layers and purposes are described clearly enough to distinguish. cg_maintain is broad and bundles several actions, which slightly increases selection ambiguity.
All tools consistently use the cg_ prefix followed by a short snake_case verb or noun (cg_write, cg_recall, cg_graph, cg_import_vault). The convention is predictable across the entire set with no mixed casing or stylistic deviations.
Twelve tools are well-scoped for a cognitive-memory graph server, covering create/read/update/forget, linking, traversal, recall, goals, reflection, maintenance, stats, and import. Each tool has a discernible role, and the count is not bloated or too thin.
The surface covers the core memory lifecycle plus maintenance, goals, reflection, import, and stats, so agents can perform most domain workflows without dead ends. Minor gaps exist, such as explicit edge deletion/unlinking and perhaps batch operations, but these are not severe for the stated purpose.
Maintenance
Related MCP Connectors
Token-efficient MCP memory for Markdown vaults. Tiered search, GraphRAG, AI memories.
Persistent memory for AI agents. Semantic search, memory graph, W3C DID identity.
Portable AI memory shared across models and harnesses - plain markdown you own.
A self-improving memory layer. Your memory, notes, tasks and goals, remembered everywhere.
Related MCP Servers
- AlicenseNot gradedqualityDmaintenanceProvides persistent long-term memory for AI assistants with tag-based retrieval, wiki-style linking, and source references, storing memories as markdown files with SQLite index.2MIT
- AlicenseAqualityDmaintenanceEnables AI agents to maintain a persistent, queryable memory stored as user-owned Markdown files, with dual-channel retrieval (FTS5 and optional semantic search) and an audited write pipeline.11MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI agents like Codex, Grok Build, and Antigravity to share persistent local memory, preferences, project context, tasks, and handoffs using Markdown and SQLite.1MIT
- AlicenseNot gradedqualityCmaintenanceGives AI agents a durable, locally hosted temporal memory built from a markdown vault you own, exposing hybrid semantic/lexical/entity search over events, threads, transcripts, entity relationships with validity dates, and patterns with confidence decay. Writes are approval-gated, so agents can draft, commit, and sync new memories while keeping every index rebuildable and nothing leaving your machine.MIT