JSPM

  • Created
  • Published
  • Downloads 48
  • Score
    100M100P100Q82144F
  • License GPL-3.0

OpenCode plugin for persistent memory using PostgreSQL + pgvector — four-layer memory architecture with topic segmentation

Package Exports

  • opcode-pg-memory
  • opcode-pg-memory/dist/index.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (opcode-pg-memory) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

opcode-pg-memory

OpenCode 长期记忆插件 — PostgreSQL + pgvector 四层记忆架构,话题隔离 · Agent 自主调用 · OmO 多 Agent 协调

npm License: GPL v3

安装

# npm 安装(推荐)
bunx opcode-pg-memory install

# 或从 GitHub
git clone https://github.com/Vbs313/opcode-pg-memory.git && cd opcode-pg-memory
.\scripts\setup.ps1

bunx install 会自动注册插件到 opencode.jsonc 并创建 /pg-memory-init 命令。

注册

~/.config/opencode/opencode.jsonc 中确认 MCP 配置:

{
  "mcp": {
    "pg-memory": {
      "type": "local",
      "command": ["bun", "path/to/opcode-pg-memory/dist/mcp-server.js"],
      "enabled": true,
      "environment": {
        "PG_HOST": "localhost", "PG_PORT": "5432",
        "PG_DATABASE": "PGOMO",  "PG_USER": "opencode",
        "PG_PASSWORD": "你的密码",
        "EMBEDDING_PROVIDER": "ollama",
        "EMBEDDING_MODEL": "qwen3-embedding:0.6b",
        "EMBEDDING_DIMENSIONS": "1024"
      }
    }
  }
}

验证:opencode --print-logs --log-level INFO | findstr "PG Memory" → 看到 Plugin initialized successfully

配置

三重优先级:MCP environment > ~/.config/opencode/pg-memory.jsonc > 默认值

// ~/.config/opencode/pg-memory.jsonc(可选)
{
  "similarityThreshold": 0.7,
  "maxMemories": 15,
  "logLevel": "info",
  "compactionThreshold": 0.85
}
Provider 模型 维度
ollama qwen3-embedding:0.6b 1024
deepseek text-embedding-v2 1536
openai text-embedding-3-small 1536

MCP 工具

recall_memory

// Agent 自主调用
recall_memory({
  "query": "数据库连接池配置",
  "caller_context": { "type": "omo_agent", "current_goal": "性能调优" },
  "retrieval_strategies": ["semantic", "graph"],
  "filters": { "tier": "project" }
})

hindsight_reflect

// 跨会话反思
hindsight_reflect({ "omo_task_id": "task_123", "aggregate": true })

架构

session_map → topic_segments → entities / observations / relations / reflections
                              → semantic_cache (HNSW 动态阈值)
                              → token_usage_log

项目结构

src/
├── index.ts        # 插件入口(chat.message + event hook)
├── config.ts       # 集中配置(文件+环境+默认值)
├── cli.ts          # CLI(install 命令)
├── types.ts        # 类型定义
├── db/init-db.ts   # 数据库初始化 + 迁移
├── hooks/          # 生命周期钩子
├── mcp/            # MCP 工具实现
├── topic/          # 话题段管理器
├── services/       # 日志 · 关键词 · 隐私过滤
├── cache/          # 语义缓存
└── omo/            # OmO 适配器
mcp-server.ts       # MCP 服务入口

文档

文档 内容
USAGE_GUIDE.md 完整使用指南 — 配置 · 工具 · 运维
PLUGIN_DOCUMENTATION.md 架构详解 — 钩子系统 · 数据库设计
AGENTS.template.md AGENTS.md 模板 — OmO directory-agents-injector 自动通知子代理

License

GPL-3.0