Back to Details

inmemoria

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

In Memoria: Persistent Codebase Intelligence

In Memoria: 持久化代码库智能

In Memoria is an MCP server that learns your codebase patterns once, then exposes that intelligence to AI agents persistently. Instead of re-analyzing code on every interaction, it maintains a semantic understanding of your architecture, conventions, and decisions.
In Memoria 是一款MCP服务器，只需学习一次你的代码库模式，就能持续为AI Agent提供这种智能能力。无需在每次交互时重新分析代码，它会持续维护对你的架构、约定和决策的语义理解。

Core Concept

核心概念

Setup → Learn → Verify → Serve. After that, AI agents query persistent intelligence without repeated parsing.
设置 → 学习 → 验证 → 提供服务。完成后，AI Agent即可查询持久化智能，无需重复解析代码。

Quick Start (5 minutes)

快速开始（5分钟）

bash
undefined
bash
undefined

1. Configure for your project

1. 为你的项目进行配置

npx in-memoria setup --interactive
npx in-memoria setup --interactive

2. Build intelligence database

2. 构建智能数据库

npx in-memoria learn ./src
npx in-memoria learn ./src

3. Verify it worked

3. 验证是否成功

npx in-memoria check ./src --verbose
npx in-memoria check ./src --verbose

4. Keep it fresh (optional but recommended)

4. 保持智能更新（可选但推荐）

npx in-memoria watch ./src
npx in-memoria watch ./src

5. Expose to agents via MCP

5. 通过MCP向Agent提供服务

npx in-memoria server
undefined
npx in-memoria server
undefined

When to Use

适用场景

✅ Use In Memoria:
  • Building long-lived AI agent partnerships (Claude, Copilot, etc.)
  • Projects where consistency across sessions matters
  • Teams wanting shared codebase intelligence
❌ Skip it:
  • One-off analysis (use 
    npx in-memoria analyze [path]
    directly)
  • Simple projects agents can read directly
✅ 推荐使用In Memoria：
  • 构建长期运行的AI Agent协作（Claude、Copilot等）
  • 跨会话一致性至关重要的项目
  • 需要共享代码库智能的团队
❌ 不建议使用：
  • 一次性分析场景（直接使用 
    npx in-memoria analyze [path]
    ）
  • Agent可直接读取的简单项目

The 5 Core Commands

5个核心命令

CommandPurposeWhen
setup --interactive
Configure exclusions, paths, preferencesFirst time only
learn [path]
Build/rebuild intelligence databaseAfter setup, major refactors
check [path]
Validate intelligence layerAfter learn, before server
watch [path]
Auto-update intelligence on code changesDuring development (optional)
server
Start MCP server for agent queriesAfter check passes
Key difference: 
learn
builds persistent knowledge. 
analyze
is one-time reporting only.
命令用途使用时机
setup --interactive
配置排除项、路径和偏好设置仅首次使用
learn [path]
构建/重建智能数据库完成设置后、重大重构后
check [path]
验证智能层执行learn后、启动server前
watch [path]
代码变更时自动更新智能开发期间（可选）
server
启动MCP服务器供Agent查询检查通过后
关键区别： 
learn
用于构建持久化知识，
analyze
仅用于一次性报告。

What Agents See

Agent可查询内容

When connected, agents can query:
  • Project structure - Tech stack, entry points, architecture
  • Code patterns - Your naming conventions, error handling, patterns used
  • Smart routing - "Add password reset" → suggests 
    src/auth/password-reset.ts
  • Semantic search - Find code by meaning, not keywords
  • Work context - Track decisions, tasks, approach consistency
连接后，Agent可查询以下信息：
  • 项目结构 - 技术栈、入口点、架构
  • 代码模式 - 命名约定、错误处理、所用模式
  • 智能路由 - 例如“添加密码重置功能”→ 推荐
    src/auth/password-reset.ts
  • 语义搜索 - 按含义而非关键词查找代码
  • 工作上下文 - 追踪决策、任务、方法一致性

Troubleshooting

故障排查

IssueFix
Learn failsVerify path is correct; check file permissions
Check reports missing intelligenceRun 
learn [path]
again
Agent doesn't see new codeIs 
watch
running? Start it: 
npx in-memoria watch ./src
Server won't startRun 
check --verbose
first; if issues, rebuild: 
rm .in-memoria/*.db && npx in-memoria learn ./src
Multiple projects conflictUse 
server --port 3001
(or different port per project)
问题解决方法
Learn执行失败验证路径是否正确；检查文件权限
Check报告缺少智能重新运行 
learn [path]
Agent无法看到新代码是否运行了
watch
？启动它：
npx in-memoria watch ./src
Server无法启动先运行
check --verbose
；若仍有问题，重建数据库：
rm .in-memoria/*.db && npx in-memoria learn ./src
多项目冲突使用
server --port 3001
（或为每个项目使用不同端口）

Performance Notes

性能说明

  • Small projects (<1K files): 5-15s to learn
  • Medium (1K-10K files): 30-60s
  • Large (10K+ files): 2-5min
If learning stalls (>10min), verify you're not indexing 
node_modules/
, 
dist/
, or build artifacts—use setup's exclusion patterns.
  • 小型项目（<1000个文件）：5-15秒完成学习
  • 中型项目（1000-10000个文件）：30-60秒
  • 大型项目（10000+个文件）：2-5分钟
如果学习过程停滞（超过10分钟），请确认未索引
node_modules/
、
dist/
或构建产物——可使用设置中的排除规则。

Key Principles

核心原则

  1. Local-first - Everything stays on your machine; no telemetry
  2. Persistent - One learning pass; intelligence updates incrementally with 
    watch
  3. Agent-native - Designed for MCP; works with Claude, Copilot, and any MCP-compatible tool
  4. Pattern-based - Learns from your actual code, not rules you define
  1. 本地优先 - 所有数据均保存在你的设备上；无遥测数据
  2. 持久化 - 仅需一次学习；通过
    watch
    实现智能增量更新
  3. 原生支持Agent - 为MCP设计；可与Claude、Copilot及任何兼容MCP的工具配合使用
  4. 基于模式 - 从你的实际代码中学习，而非你定义的规则

Deployment Pattern (3 terminals)

部署模式（3个终端）

bash
undefined
bash
undefined

Terminal 1: One-time setup

终端1：一次性设置

npx in-memoria setup --interactive npx in-memoria learn ./src npx in-memoria check ./src --verbose
npx in-memoria setup --interactive npx in-memoria learn ./src npx in-memoria check ./src --verbose

Terminal 2: Keep intelligence fresh

终端2：保持智能更新

npx in-memoria watch ./src
npx in-memoria watch ./src

Terminal 3: Expose to agents

终端3：向Agent提供服务

npx in-memoria server
npx in-memoria server

Now agents (Claude, Copilot, etc.) have persistent codebase context

现在Agent（Claude、Copilot等）即可获取持久化代码库上下文


See [GitHub](https://github.com/pi22by7/In-Memoria) for full API docs and agent integration examples.

完整API文档及Agent集成示例请查看[GitHub](https://github.com/pi22by7/In-Memoria)。