frazik-main/amplifygpt

Java library to amplify your project with Auto-GPT like features

AmplifyGPT

Customizable Auto-GPT library for Java

AmplifyGPT is a Java library version of Auto-GPT with customization and i18n support in mind. It can help you easily incorporate Auto-GPT like functionality into your new or existing project.

Feature roadmap

• 🧠 Amplify JVM based project: Make your new or existing project more intelligent.
• ⚡ Customization: Customization and extensibility is the core of AmplifyGPT.
• 🌐 Internationalization: AmplifyGPT is i18n ready. You can use it in any language you want.
• ➿ GPT 3.5 and GPT 4 support: AmplifyGPT supports both GPT 3.5 and GPT 4.
• 🗃️ Multi-model support: Architected to supports multiple models (GPTx, Bard, LLaMA, StableLM ...).
• ⌚ Prompt optimization: Prompt optimization to consume fewer tokens.
• 🌀 Support Java 1.8 and above: For maximum compatibility.

Installation

• Clone the repository
• Build the project using gradle

gradle build

• Add the jar file to your project

Running examples

• Run the examples

gradle run

• Export Open AI key
• To run examples you need to export Open AI key

export OPENAI_API_KEY=<your-openai-api-key>

Requirements

• Java 1.8+
• An OpenAI API Key
• Windows/Linux/Mac
• Google API key (optional)

Want to contribute?

• We need A LOT of Help!
• You want to contribute to the project?
• Please open an issue or a PR if you'd like to contribute.

amplifygpt's Issues

More tools need to be defined

Let's define which tools we are going to implement.
We need tools like running code, running new agents, etc.

Use AmplifyGPT to create description and explanation that will serve as pitch for contributors

Make chat work

Right now, we have basic chat working. We need to ensure chat looping is working.

There is opportunity to introduce Prompt object

Try re-factor by introducing Prompt object which will serve as prompt generation for various prompts types: Tool, header, etc. Maybe there is opportunity to create hierarchical structure.

Chat results are not consistent.

Chat results are not always consistent. Maybe this is not possible to solve, let's investigate.

Extract strings to resources

Currently prompts Strings are sprinkled through code base, we need to collect them for easy translations purposes.

Introduce new tests

We need unit tests for more test cases.

Expand header prompt

Expand the first header prompt with a list of commands and a response format.

Use AmplifyGPT to update Readme.md

Redundant OpenAI api call

Setup workflows

General code cleanup

Let's cleanup the code. It will be general sweep of low hanging fruits.

Introduce token optimization with document similarity

Make summarizer work

Summarizer is not working, make it work.

Initialize agent tools

When agent is being created, initialize list of availible tools.

Fix google search

stagingTool in Agent.java have <String, Object> type, but sometimes it behaves as <String, String)>

Incorporate subAgents

Interesting option is to have subAgents that can be forked from original agent.

Optimize agent name through code base

Agent name is passed all over the place. We need one place from where agent name will be set.

Update package (package is using obsolete name)

Fix LocalMemory funcionality

When memory.add is called it will create embeddings, but than it freezes on the line:

•                                                                       INDArray emb = Nd4j.create(embeddings);
  

It is maybe some problem with Nd4j library

Make code base more maintainable/readable

There are several opportunities to make code-base batter.

1. Decrease responsibility of Agent class
2. Better and wider usage of Prompt and don't return exist format (String, Map<String, String>) until needed for OpeanAI API
3. Re-factor response to better and re-usable format
4. Decrease cognitive complexity in Agent object
5. stagingTool in Agent.java have <String, Object> type, but sometimes it behaves as <String, String)>
6. Each tool should handle it's own prompt creation

Full prompt is not formatted correctly

We are missing new lines.
One prompt section necessary have numbers.
Seed is not used at all.

Token limit should be 4000, not 3900

Chat doesn't work on basic input

Chat doesn't continue after first tool execution replay.

Fix stagingTool in Agent.java

stagingTool in Agent.java have <String, Object> type, but sometimes it behaves as <String, String)>
We need to investigate this and fix it.

Update README

README need to be more approachable and descriptive