update kb design, base on objfs and knowled graph.
This commit is contained in:
@@ -0,0 +1,34 @@
|
||||
# agent memory
|
||||
|
||||
## memory的基本形式
|
||||
memory的基本形式上是 topic+内容
|
||||
topic用一个有意义的路径表示 /xxx/xxx/xxx (有点类似脑图的逻辑,可以通过逐级展开遍历浏览所有的memory)
|
||||
同一个memory可以被多个路径指向
|
||||
内容则是一个json文件
|
||||
|
||||
|
||||
## Agent 使用memory的
|
||||
1. 根据当前会话的主题,尝试在known_info中加载必要的memory
|
||||
2. 提供memory的 list/查询 函数, 允许agent在必要的时候 list / 查询memory
|
||||
该使用逻辑的本质和kb查询逻辑很像
|
||||
|
||||
## Agent 更新/创建memory
|
||||
1. 在任何llm process的过程中,agent都可以用写文件的形式创建memory
|
||||
2. 更新memory通常是一个专门的 self-think过程,agent此时会用某种模式整理自己所有的logs和memory,并对memory进行更新、创建、删除
|
||||
该更新逻辑与Agent 与KB的Self-learning逻辑很像。但根据log->summary的过程基本上是 self-think独有的
|
||||
|
||||
## 实现逻辑
|
||||
基本思路:
|
||||
1. 核心API是一组通用的文件操作API(有些场景可以是只读的) + 一组特化的对象查询API
|
||||
路径->Object,Object中包含ObjectId等信息
|
||||
ObjectId->Object
|
||||
Object一定是一个json,里面包含可以打开原始文件的路径(fileId)
|
||||
2. 通过一组文件系统描述来引导Agent操作特定文件
|
||||
3. 通过一组搜索API来引导Agent操作特定文件
|
||||
|
||||
对象查询API,基本思路是
|
||||
|
||||
ObjectId->Object
|
||||
|
||||
|
||||
|
||||
@@ -18,6 +18,7 @@ Let's start by introducing the two important processes.
|
||||
Note that the dependency check during installation allows for the missing packages to be installed into the current environment.
|
||||
|
||||
# Some Basic Concepts
|
||||
|
||||
- ***env***:A target environment consisting of a series of configuration files, where packages can be loaded/installed.
|
||||
- ***pkg***:A Package(pkg) is either a folder or a file that serves the same purpose as a folder (such as zip, iso, etc.).
|
||||
- ***pkg_name***:A unique string used to label a package. It's usually a readable package name, but can also include the version number or even the ContentId.
|
||||
|
||||
@@ -0,0 +1,15 @@
|
||||
# prompts
|
||||
|
||||
四个循环
|
||||
1. 立刻处理消息+深度思考整理循环
|
||||
Process <-> Self Thinking
|
||||
|
||||
2. 任务迭代完整循环致力于完成所有的待完成任务(不一定是成功完成)
|
||||
Task -> Todo -> Check
|
||||
|
||||
3. 知识库整理
|
||||
New Knowledge -> Self-Learning
|
||||
使用知识库的时机?是否有quick process和deep thing的区别?
|
||||
|
||||
4. Self-Improve
|
||||
根据四元组:输入,提示词,输出,上级意见 (可选),对提示词进行改进
|
||||
Reference in New Issue
Block a user