Releases: AgentEra/Agently
v3.0.4
New Feature:
Agent Component - Search!
How to use:
agent\
.set_role("As a preschool teacher, I'm here to turn complex and hard-to-understand knowledge into stories that little ones can understand. Even though it's a story, it's essential to ensure the accuracy and truthfulness of the information.")\
.toggle_component("Search", True)\
.instruct("If the search results or extra information contain a lot of content, try to organize it into multiple structured story segments as much as possible.")\
.input("Why the sky is blue?")\
.start()Other Important Updates:
- Core Update: Support modify prefix stage call orders, you can modify it by editing config.ini file in plugins/agent_component dir.
新能力：
Agent组件 - 搜索
如何使用：
agent\
.set_role("你是一个幼儿教师，你需要将晦涩难懂的专业知识理解之后转化成小孩子能听懂的故事讲给用户听，注意，虽然是讲故事，但是要保证专业知识的准确真实")\
.toggle_component("Search", True)\
.instruct("如果搜索结果中包含较多内容，请尽可能将这些内容有条理系统地转化成多段故事")\
.input("天为什么是蓝色的")\
.start()其他重要更新：
- 框架核心升级： 支持在prefix阶段设置组件的执行顺序，你可以通过修改plugins/agent_component文件夹里的config.ini文件来调整顺序
Full Changelog: Maplemx/Agently@v3.0.3...v3.0.4
v3.0.3
Introducing Agently 3.0 - An AI Agent Native Development Framework
Agently is a development framework that helps developers build AI agent native application really fast.
You can use and build AI agent in your code in an extremely simple way.
You can create an AI agent instance then interact with it like calling a function in very few codes.
In AI agent native application, we put an AI agent instance into our code, then we ask it to execute / to solve the problem with natural language or natural-language-like expressions.
"Ask-Get Response" takes place of traditional "Define Problem - Programme - Code to Make It Happen".
Visit this link, it is a colab document, you can run code demos online and see how easy develop using agent can be done.
Also, Agently 3.0 re-design the plugins mechanism that brings better support for plugin developers to participate plugin development contribution.
Click here to read plugin code in framework as examples to see how easy to develop plugins especially agent component plugins.
If you like our work, please star our project.
v2.0.5-Python
Bug Fixed:
- Fixed when try to fix JSON format in workflow, process sometimes report Error: can not find Type 'customize'; https://github.com/Maplemx/Agently/issues/23
- Fixed a bug when request XunFei Spark model, the result is None; https://github.com/Maplemx/Agently/issues/20
v2.0.5 is the last stable version of Agently v2.0(Python)
We will soon publish the brand new Agently v2.1(Python) agent framework with a completely new architectural design and refactoring.
v2.0.4-python
新功能：
- 支持
百度文心模型库和讯飞星火大模型； - 提供了一个非常易用的一键启动shell脚本 ，用于启动Agent DEMO，你可以参看blueprint目录了解如何使用它启动自己的Agent蓝图；
- 提供了一个能够分析用户意图并给出更恰当回答的Agent蓝图：SmartOne；
- 在Session上新增了async_start()方法，以更方便您在async环境中编写协程；
New Features:
- Support Baidu WenXin model workshop and XunFei Spark model;
- One-Command Shell Demo Starter that super easy to use and you can run your own agent blueprint by it;
- Agent Blueprint Example: SmartOne;
- new method Session.async_start() to help you coding with asyncio coroutine;
变更：
- 调整了work node中request节点的请求过程事件表达格式，以<阶段:事件>来进行命名；
- 调整了request节点的文件结构，让编写不同模型的请求支持更加简单；
- 移除了"register_response_handler"节点，新增“register_response_suffix"”节点，用于在最终向应用开发层抛出
delta和done事件数据之前，对数据进行处理； - 完善了requirements.txt文件。
Update:
- Change event expression in request work node, use Stage:Event to make event clearly to be understand;
- Change file organize structure of llm request to make it easier to add more llm request support;
- remove work node "register_response_handler" and add a new work node "register_response_suffix"
- Update requirements.txt.
Full Changelog: Maplemx/Agently@v2.0.1-python...v2.0.4-python
v2.0.1-python
Agently is now available in Python!
☄️ Only a few lines of code brings you an LLM agent worker in code to help you correct JSON string format error, generate API request parameters from natural languages, etc.
🎭 Easy ways to manage agent role settings, switch models, control status and context.
🧩 Arrange your agents' own Workflow and modify your own Work Nodes' working logic by code. Test your own chain of thoughts to make your agent smarter.
👥 Share your own designed agent through Blueprint to community.
☄️ 几行代码开箱即用，在你的代码里就可以拥有一个agent帮你改错误的数据格式，生成接口调用数据
🎭 轻松便捷的Agent设定管理方法，轻松切换模型，管理人物设定、状态数值和上下文
🧩 通过工作节点（work node）和工作流(workflow)来编排你的Agent行为，让它通过思维链或是其他方式变得更聪明
👥 通过蓝图的方式进行分享，让别人能够快速用上你设计的独特Agent
v1.1.3
This version fixed a serious when making streaming request and this is a stable version to use for now.
Agently develop team will move on to develop brand-new version v2.0.0 and hope to release as soon as possible.
该版本修复了Streaming请求时会发生的严重错误，目前已经是一个相对稳定的版本，能够帮助大家理解Agently的理念，并进行面向Agent的开发，验证基于Agent的各种LLM协同想法。
目前Agently开发团队正在努力进行全新的v2.0.0版本开发，v2.0.0版本将提供更加层次清晰和健壮的框架主体以及支持社区生态共同定制开发的协作模式。
我们希望这个版本尽快和大家见面。祝大家玩得愉快。
v1.1.0: Use Skills to Enhance Your Agent
What's New?
Do you know the full name of GPT is "Generatvie Pre-trained Transformer"? From "Pre-trained" we know, LLM is not that kind of model can keep up current news and events. As we all well-known that when GPT3.5 just born, its knowledge can just reach up to the year 2021. Because it's "Pre-trained", before its next training, it will stay to its status.
If we want LLM models to catch up in some areas, what can we do? One idea could be to equip the LLM-based agents with some skills.
In fact, to catch up is only a small thing that skills can enchane the agent. Just think, what can an agent do when it can browse webpages or write down some to-dos into the list for you for real.
Agently provides a convenient way to do that.
全新功能：使用技能（Skills）来增强你的Agent
GPT的全称是“Generatvie Pre-trained Transformer”，翻译过来，是基于Transformer的用于生成的预训练模型。从“预训练”三个字我们可以知道，目前的大语言模型并不是那种可以实时更新自己的信息的模型。在GPT 3.5刚刚问世的时候（那是2022年的年末），它的知识储备，也仅仅储备到了2021年，也就是说，2021年之后世界发生的实际变化，它并不知道。“预训练”也就意味着，在下一次训练添加更新的事实语料之前，模型只能保持自己之前训练完成的状态。
如果我们希望我们使用的基于大语言模型工作的Agent能够在某些方面能够跟上世界的变化，我们能做什么呢？或许，给Agent添加一些技能（Skills）让它能够和真实世界发生交互，会是一个好主意。
HOW TO INSTALL?
npm：npm install agently
yarn：yarn add agently
HOW TO USE?
🤵 Agently is a framework helps developers to create amazing LLM based applications.
🎭 You can use it to create an LLM based agent instance with role set and memory easily.
⚙️ You can use Agently agent instance just like an async function and put it anywhere in your code.
🧩 With the easy-to-plug-in design, you can easily append new LLM API/private API/memory management methods/skills to your Agently agent instance.
⚠️ Notice: Agently is a node.js package only works on the server-side.
🥷 Author: Maplemx | 📧 Email: [email protected] | 💬 WeChat: moxinapp
⭐️ Star this repo if you like it, thanks!
🤵 Agently是一个希望帮助大语言模型（LLM）应用开发者们制作出超棒的大语言模型应用（LLM Based Applications）的轻量级框架
🎭 你能够使用Agently快速而轻松地创建并管理基于大语言模型的Agent实例，并管理他们的人设和记忆，这将让客服机器人、角色扮演机器人、游戏用Agent的构造和管理更方便
⚙️ 你可以把Agently创建的Agent以及Session像一个异步函数（async function）一样使用，这将让基于大语言模型能力的自动化工作流构造更轻松，你甚至可以沿用原有的业务代码，在其中部分需要NLP算法、复杂推理或人工操作的环节，尝试把Agently提供的Agent和Session当做一个异步函数，几乎无缝地加入到代码的业务流程中
🧩 Agently在设计时考虑了对主要请求流程中节点部件的可更换性，你可以轻松地更换或定制它们，例如：添加新的LLM模型请求方法，更换私有/转发的模型请求API地址，调整Agent记忆管理方法，定制自己的模型消息解析方案等
🔀 Agently提供的独特的针对一次请求中的流式消息（Streaming Message）的消息分块及多下游分发管理方案，能够让你在接收流式消息时，一方面保留了大语言模型通过流式消息的方式带来的高速反馈敏捷性优点，另一方面又能在一次请求中做更多的事情
⚠️ 注意：Agently适用于Node.js的服务端而不是网页前端
🥷 作者: Maplemx | 📧 Email: [email protected] | 💬 微信: moxinapp
⭐️ 如果您觉得这个项目对您有帮助，请给项目加星，感谢您的肯定和支持！
v1.0.0
HOW TO INSTALL?
npm：npm install agently
yarn：yarn add agently
HOW TO USE?
🤵 Agently is a framework helps developers to create amazing LLM based applications.
🎭 You can use it to create an LLM based agent instance with role set and memory easily.
⚙️ You can use Agently agent instance just like an async function and put it anywhere in your code.
🧩 With the easy-to-plug-in design, you can easily append new LLM API/private API/memory management methods/skills to your Agently agent instance.
🥷 Author: Maplemx | 📧 Email: [email protected] | 💬 WeChat: moxinapp
⭐️ Star this repo if you like it, thanks!
🤵 Agently是一个希望帮助大语言模型（LLM）应用开发者们制作出超棒的大语言模型应用（LLM Based Applications）的轻量级框架
🎭 你能够使用Agently快速而轻松地创建并管理基于大语言模型的Agent实例，并管理他们的人设和记忆，这将让客服机器人、角色扮演机器人、游戏用Agent的构造和管理更方便
⚙️ 你可以把Agently创建的Agent以及Session像一个异步函数（async function）一样使用，这将让基于大语言模型能力的自动化工作流构造更轻松，你甚至可以沿用原有的业务代码，在其中部分需要NLP算法、复杂推理或人工操作的环节，尝试把Agently提供的Agent和Session当做一个异步函数，几乎无缝地加入到代码的业务流程中
🧩 Agently在设计时考虑了对主要请求流程中节点部件的可更换性，你可以轻松地更换或定制它们，例如：添加新的LLM模型请求方法，更换私有/转发的模型请求API地址，调整Agent记忆管理方法，定制自己的模型消息解析方案等
🔀 Agently提供的独特的针对一次请求中的流式消息（Streaming Message）的消息分块及多下游分发管理方案，能够让你在接收流式消息时，一方面保留了大语言模型通过流式消息的方式带来的高速反馈敏捷性优点，另一方面又能在一次请求中做更多的事情
🥷 作者: Maplemx | 📧 Email: [email protected] | 💬 微信: moxinapp
⭐️ 如果您觉得这个项目对您有帮助，请给项目加星，感谢您的肯定和支持！
v0.0.1
Version v0.0.1 of Agently is a tools I develop for speeding the feasibility tests of my language model based application idea. I did not think too much to make it work in product environment. So there still are many functions to work on like streaming the reply, better memory management, etc.
But I notice some people have already use this version through npm install...
Because the next version(v1.0.0) will be publish soon and totally changed. I hope that I can keep a version snapshot by release for those who have already use Agently v0.0.1.