catalog/repos/5kahoisaac--opencode-historian.md

AI对话持久记忆插件

GitHub: 5kahoisaac/opencode-historian | Gitea: mc-skills/5kahoisaac--opencode-historian

中文摘要

为OpenCode智能代理提供跨会话持久记忆能力，支持语义搜索、记忆分类存储（决策/偏好/问题/经验等），以Markdown格式保存并通过QMD索引，让AI代理在多次对话间记住项目上下文。

标签

持久记忆 AI代理 OpenCode 语义搜索 知识管理

---

README 原文

Historian gives your AI agent persistent memory across conversations.

Sync · Recall · Remember · Forget · Compound

---

OpenCode Historian

Persistent memory for OpenCode agents, powered by QMD.

Historian helps your agent remember decisions, preferences, learnings, and project context across sessions. It stores memories as markdown, indexes them with QMD, and exposes tools for remembering, recalling, forgetting, and syncing memory.

What You Get

• Persistent memory across conversations
• Semantic search over saved memories
• Built-in memory types for decisions, issues, learnings, and preferences
• Markdown-based storage in your repo
• A bundled historian agent and memory tools
• Optional Serena MCP support for code navigation

Prerequisites

• Bun 1.3.9+
• QMD installed globally

Install QMD:

npm install -g qmd
# or
bun install -g qmd

Install the Plugin

Add the plugin to your opencode.json:

{
  "plugins": ["opencode-historian"]
}

That is enough to register the plugin, the bundled historian agent, and the memory tools.

Install the mnemonics Skill

The plugin also includes a mnemonics skill that teaches agents how to use the @historian subagent effectively.

Recommended install:

npx skills add https://github.com/5kahoisaac/opencode-historian/tree/main/src --skill mnemonics

After installing it, agents can load mnemonics for guidance on memory types, when to use @historian, and how to store or recall project knowledge correctly.

Quick Start

Once the plugin is enabled, talk to the historian agent in natural language.

Save information

"Remember that we're using PostgreSQL for the database"

"Save this: we decided on JWT tokens with 24-hour expiry"

"Note that the API rate limit is 100 requests per minute"

Recall information

"What did we decide about authentication?"

"Do we have any known issues?"

"What are my preferences for this project?"

Historian will:

• classify memories by type
• tag them for retrieval
• index them for semantic search
• keep them in a git-friendly markdown format

Memory Types

Historian ships with these built-in memory types:

Type	Use For
architectural-decision	System architecture choices
design-decision	UI/UX decisions
learning	Lessons and discoveries
user-preference	User preferences
project-preference	Team conventions
issue	Known problems
context	General context (default)
recurring-pattern	Reusable patterns
conventions-pattern	Coding standards

Configuration

Optional config file:

.opencode/opencode-historian.json

Example:

{
  "appendPrompt": "Focus on API design decisions.",
  "memoryTypes": [
    {
      "name": "api-endpoint",
      "description": "API endpoint decisions"
    }
  ],
  "disabledMcps": ["serena"]
}

Options

Option	Default	Description
model	-	Model used by the historian agent
temperature	0.3	Response creativity
appendPrompt	-	Additional instructions appended to the historian prompt
memoryTypes	-	Custom memory types to add alongside the built-ins
autoCompound	true	Automatically merge new learnings into existing memories when appropriate
disabledMcps	-	Bundled MCPs to disable, for example ["serena"]

Storage

Memories are stored as markdown files under .mnemonics/ in your project root:

.mnemonics/
├── architectural-decision/
├── design-decision/
├── learning/
└── ...

Benefits:

• human-readable
• easy to version with git
• easy to inspect or edit manually

Included Tools

The plugin registers these memory tools:

• memory_remember
• memory_recall
• memory_forget
• memory_list_types
• memory_sync

When to Use the Skill vs. the Tools

• Use the plugin to make memory available inside OpenCode
• Use the mnemonics skill to teach agents how to use @historian well
• Use the memory tools when you want direct programmatic memory operations

In short: the plugin gives you capability, and the skill gives agents better judgment about how to use it.

License

This project is licensed under the MIT License.