
EchoVault Memory
io.github.go-ports由 GitHub 账号 go-ports 发布v0.3.3更新于 Oct 24, 2026
Local-first memory for coding agents; stores decisions, bugs, and context across sessions.
发布方自己的服务:Memory.fit
概览
面向编码代理的本地优先记忆服务器,把决策、缺陷与上下文以 Markdown 保存在本机,并支持跨会话保存、搜索和重新加载。
- 可以做什么
- 保存记忆,包括标题、原因、影响、标签和分类,并支持结构化详情模板或详情文件,之后可按 ID 或前缀删除。用关键词与可选语义向量混合搜索记忆,并获取某条结果的完整详情。在会话开始时加载既有上下文,列出当前项目的记忆和会话文件。让 Claude Code、Cursor、Codex 与 OpenCode 共用一个记忆库,并通过命令行管理记忆目录和嵌入配置。
- 适用场景
- 当你希望编码代理把决策、缺陷历史和项目上下文从一个会话带到下一个会话,又不依赖云服务时,适合安装。若你同时使用多个代理并希望它们共享一个可用 Markdown 浏览的记忆库,也很有用。
- 运行要求
- 以 stdio 作为本地进程运行;Go 版本是单一静态二进制文件,从源码构建需要 C 编译器,因为要链接 SQLite 库。需先初始化记忆库并安装代理配置。环境变量 MEMORY_HOME 指定存储目录,应指向持久化存储。语义搜索可选依赖本地 Ollama,或在配置文件中填写 OpenAI 或 OpenRouter 的 API 密钥。
安装
在 SourceWeft 中
- 打开 控制台中的 EchoVault Memory,将其添加到工作区。
- 为需要使用其工具的对话启用该服务。
Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。
其他 MCP 客户端
参照 仓库 中的启动说明。
README
EchoVault — Go
Local memory for coding agents. Your agent remembers decisions, bugs, and context across sessions — no cloud, no API keys, no cost.
Install · Features · How it works · Commands
This is the Go port of EchoVault. It is a single static binary with no Python runtime dependency. The vault format and MCP interface are fully compatible with the Python version — you can switch between them without losing any memories.
Features
Works with 4 agents — Claude Code, Cursor, Codex, OpenCode. One command sets up MCP config for your agent.
MCP native — Runs as an MCP server exposing memory_save, memory_search, and memory_context as tools. Agents call them directly — no shell hooks needed.
Local-first — Everything stays on your machine. Memories are stored as Markdown in ~/.memory/vault/, readable in Obsidian or any editor.
Zero idle cost — No background processes, no daemon, no RAM overhead. The MCP server only runs when the agent starts it.
Hybrid search — FTS5 keyword search works out of the box. Add Ollama or OpenAI for semantic vector search.
Secret redaction — 3-layer redaction strips API keys, passwords, and credentials before anything hits disk. Supports explicit <redacted> tags, pattern detection, and custom .memoryignore rules.
Cross-agent — Memories saved by Claude Code are searchable in Cursor, Codex, and OpenCode. One vault, many agents.
Obsidian-compatible — Session files are valid Markdown with YAML frontmatter. Point Obsidian at ~/.memory/vault/ and browse your agent's memory visually.
Install
Pre-built binary
Download the latest release for your platform from the releases page and place the binary somewhere on your $PATH.
Build from source
CGO required. The binary links against
go-sqlite3andsqlite-vec, so a C compiler (gcc/clang) must be present. On macOS:xcode-select --install. On Debian/Ubuntu:apt install build-essential.
First run
That's it. memory setup installs the MCP server config automatically.
By default the config is installed globally. To install for a specific project:
Configure embeddings (optional)
Embeddings enable semantic search. Without them, you still get fast keyword search via FTS5.
Generate a starter config:
This creates ~/.memory/config.yaml with sensible defaults:
What each section does:
embedding— How memories get turned into vectors for semantic search.ollamaruns locally;openaiandopenroutercall cloud APIs.nomic-embed-textis a good local model for Ollama.context— Controls how memories are retrieved at session start.autouses vector search when embeddings are available, falls back to keywords.topup_recentalso includes recent memories so the agent has fresh context.
For cloud providers, add api_key under the provider section. API keys are redacted in memory config output.
Configure memory location
By default, EchoVault stores data in ~/.memory.
You can change that in two ways:
MEMORY_HOME=/path/to/memory(highest priority, per-shell/per-process)memory config set-home /path/to/memory(persistent default)
Useful commands:
memory config shows both memory_home and memory_home_source (env, config, or default).
The --memory-home global flag overrides everything for a single invocation:
Usage
Once set up, your agent uses memory via MCP tools:
- Session start — agent calls
memory_contextto load prior decisions and context - During work — agent calls
memory_searchto find relevant memories - Session end — agent calls
memory_saveto persist decisions, bugs, and learnings
The MCP tool descriptions instruct agents to save and retrieve automatically. No manual prompting needed in most cases.
You can also use the CLI directly:
For long details, use --details-file notes.md. To scaffold structured details automatically, use --details-template.
How it works
- Markdown vault — one file per session per project, with YAML frontmatter
- SQLite index — FTS5 for keywords, sqlite-vec for semantic vectors
- Compact pointers — search returns ~50-token summaries; full details fetched on demand
- 3-layer redaction — explicit tags, pattern matching, and
.memoryignorerules
Supported agents
All agents share the same memory vault at your effective memory_home path (default ~/.memory/). A memory saved by Claude Code is searchable from Cursor, Codex, or OpenCode.
Commands
Global flags
Uninstall
To also remove all stored memories: rm -rf ~/.memory/
Privacy
Everything stays local by default. If you configure OpenAI or OpenRouter for embeddings, those API calls go to their servers. Use Ollama for fully local operation.
License
MIT — see LICENSE.
来源:README.md,提交 cc0802b
工具
0版本历史
1- v0.3.3最新Sep 16, 2026
评分
已满足适用检查项的 50%
该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。
- 近 30 天内可连通25 分
请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。
- 已声明认证方式10 分
请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。
- 已列出工具15 分
请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。
- 描述必需15 分
- 图标10 分
- 文档15 分
- 许可证10 分
其他 Memory 集成

Memory.fit
fit.memory官方发布者
托管式 Memory.fit 服务器,让 AI 读取你的训练、睡眠、恢复与饮食数据,并写回每周计划、每日目标和饮食计划。
pg.ddx.io: the PostgreSQL community's memory
io.ddx.pg官方发布者
pg.ddx.io: the PostgreSQL community's memory 的托管 MCP 服务,由 pg.ddx.io 发布;无需登录;Streamable HTTP;29 个工具。
var.gg Naming Memory
gg.var官方发布者
远程 MCP 服务器,可根据领域概念推荐并检索韩、日、中、越、英多语言的代码标识符名称,并提供 ERD 结构查询。

Counterparts
ai.counterparts
面向 Claude Code 与 Claude Desktop 的本地优先长期记忆:助手自行写下关于你和项目的笔记,并在之后的会话中回忆起来。

LAHZE - Persian AI Memory Platform
由 GitHub 账号 pedramgholizadeh 发布
以波斯语为先的地理记忆平台:发现地点,起草并发布仅限波斯语的记忆。AI 翻译