ROADMAP 2024

[Undertone0809]

The issue is for the development plan of promptulate v2.0.0. The In Development status indicates that there is ongoing progress on the feature, while Planned Development signifies the functionalities intended to be completed in this version but have not yet started. If you are interested in any of the points, you are welcome to create an issue about that content with a simple proposal, describing your general implementation ideas for discussion.

Please note:

1. If you want to develop a certain feature, I highly recommend you start a new issue mentioning that content to avoid duplicate development efforts. On the other hand, issues provide a channel for developers to communicate.
2. If someone proposes to develop a content listed under Planned Development, I will move it to In Development.
3. There are some other development plans in the official documentation that are not yet included in the current version's development plan. If you see a feature you would like to work on, we would be very happy to add it to the development plan for this version!

If you have more interesting feature ideas, you are very welcome to suggest them, as well as submit related issues and pull requests!

Completed

• Support for Wenxin Yiyan model #22
• IoT control support #64
• Shell Tool support #55 #62
• Compatibility with langchain tools #67
• Compatibility with huggingface tools #67
• Transition from BaseTool to Tool type
• Build vercel-promptulate-flask https://github.com/Undertone0809/vercel-promptulate-flask
• • Compatibility with huggingface models
• Unit testing and black formatting checks during pull requests

In Development

• Providing Chroma integration solution
• Providing various external data access solutions, such as documents, URLs, etc.
• Using OpenAI for GPT capability driving
• Providing lazy loading for all modules

Planned Development

Agent

• Support for more types of Agents

Tool

• File read/write tools
  • Create/delete file Tool
  • File write/append Tool
  • File reading tools
• Web scraping tools
• Building more types of tools
  • Database query tools: Construction of SqlToolKit
  • Reference: SqlDatabaseTookKit
  • Integration with more types of search APIs, such as Google
• Few-shot support for Tool descriptions
• Reference construction of API-style tool extensions, seamlessly interfacing with existing tools
  • References
  • Connecting massive APIs with large models: Gorilla
  • https://arxiv.org/pdf/2305.15334.pdf

LLM

• Adaptation to more LLMs, such as current various open-source models
  • ChatGLM
  • ChatGLM2-6B
  • iFlytek Spark Large Model

Memory

• Memory storage for Agent, llm related configuration information
• Memory compatibility with relational database storage

Others

• Improve documentation, add new sections, provide knowledge on LLM, prompt techniques
• Develop a simple feature demonstration server based on gradio (to be discussed)
• Build a plugin marketplace
  • Build plugin templates
  • Integrate third-party plugins quickly via command line
• Build multi-Agent scheduling model
• Provide proxy configuration for DuckDuckGo
• Build a custom Hook system
• Build streaming transmission
• Explore multimodal capabilities: text-to-image, integration with Huggingface models
• Output parser

Discuss

• Should the role of BaseFramework and the functionality of presets be removed, or is there a better alternative?
• Should the embedding method of Tree of Thoughts be introduced as a mixin? Further discussion is needed.
• Is it possible to build an adapter for langchain, thereby connecting to the langchain ecosystem?
• Is it feasible to build an LLM's huggingface, offering a more streamlined and elegant way to quickly deploy a large language application model?

Translate into English version.

[Undertone0809]

Next stage, we are going to focus on promoting Agent module and pne.chat() module

[Undertone0809]

Now we use Github Kanban to manage promptulate roadmap. https://github.com/users/Undertone0809/projects/3