Agentica: Build AI Agents

Agentica: Effortlessly Build Intelligent, Reflective, and Collaborative Multimodal AI Agents!

📖 Introduction

Agentica can build AI agent, which have the components of planning, memory, and tool use.

Agent Components

Planning: task decomposition, plan generation, reflection
Memory: short-term memory (prompt implementation), long-term memory (RAG implementation)
Tool use: function call capability, call external API to obtain external information, including current date, calendar, code execution capability, access to dedicated information sources, etc.

Agentica Workflow

Agentica can also build multi-agent systems and workflows.

Planner: responsible for LLM to generate a multi-step plan to complete complex tasks, generate interdependent "chain plans", and define the output of the previous step that each step depends on
Worker: accepts the "chain plan", loops through each subtask in the plan, and calls tools to complete the task, which can automatically reflect and correct errors to complete the task
Solver: The solver integrates all these outputs into the final answer

🔥 News

[2024/12/29] v0.2.3: Added support for ZhipuAI API calls, including free models and tool usage. See Release-v0.2.3

[2024/12/25] v0.2.0 version: Supports multimodal models, input can be text, pictures, audio, video, upgrade Assistant to Agent, Workflow supports disassembly and implementation of complex tasks, see Release-v0.2.0

[2024/07/02] v0.1.0 version: Implemented Assistant based on LLM, can quickly use function call to build a large language model assistant, see Release-v0.1.0

😊 Features

Agentica is a tool for building agents with the following features:

Agent Composition: Quickly compose agents with simple code, supporting Reflection, Plan and Solve, RAG, Agent, Multi-Agent, Multi-Role, and Workflow functionalities.
Custom Prompts: Agents support custom prompts and various tool calls.
LLM Integration: Supports APIs from multiple large model providers such as OpenAI, Azure, Deepseek, Moonshot, Anthropic, ZhipuAI, Ollama, and Together.
Memory Capabilities: Includes short-term and long-term memory functionalities.
Multi-Agent Collaboration: Supports team collaboration with multiple agents and roles.
Workflow Automation: Automates complex workflows by breaking down tasks into multiple agents, such as investment research, news article writing, and technical tutorial creation.
Self-Evolving Agents: Agents with reflective and memory-augmented abilities that can self-evolve.
Web UI: Compatible with ChatPilot, supporting web-based interaction through popular front-end frameworks like open-webui, streamlit, and gradio.

Install

pip install -U agentica

or

git clone https://github.com/shibing624/agentica.git
cd agentica
pip install .

Getting Started

1. Install requirements

git clone https://github.com/shibing624/agentica.git
cd agentica
pip install -r requirements.txt

2. Run the example

# Copying required .env file, and fill in the LLM api key
cp .env.example ~/.agentica/.env

cd examples
python web_search_moonshot_demo.py

Copy the .env.example file to ~/.agentica/.env and fill in the LLM API key (optional DEEPSEEK_API_KEY, MOONSHOT_API_KEY, OPENAI_API_KEY, etc.). Or you can set the environment variable directly.
```
export OPENAI_API_KEY=your_openai_api_key
export SERPER_API_KEY=your_serper_api_key
```
Build and run an Agent using agentica:

Automatically call the Google search tool, example examples/11_web_search_moonshot_openai.py

from agentica import Agent, OpenAIChat, SearchSerperTool

m = Agent(model=OpenAIChat(id='gpt-4o'), tools=[SearchSerperTool()], add_datetime_to_instructions=True)
r = m.run("Where will the next Olympics be held?")
print(r)

Web UI

shibing624/ChatPilot is compatible with agentica and can be interacted with through a Web UI.

Web Demo: https://chat.mulanai.com

git clone https://github.com/shibing624/ChatPilot.git
cd ChatPilot
pip install -r requirements.txt

cp .env.example .env

bash start.sh

Examples

Example	Description
examples/01_llm_demo.py	LLM Q&A Demo
examples/02_user_prompt_demo.py	Custom user prompt Demo
examples/03_user_messages_demo.py	Custom input user messages Demo
examples/04_memory_demo.py	Agent memory Demo
examples/05_response_model_demo.py	Demo of responding in a specified format (pydantic's BaseModel)
examples/06_calc_with_csv_file_demo.py	Demo of LLM loading CSV files and performing calculations to answer questions
examples/07_create_image_tool_demo.py	Demo of creating an image tool
examples/08_ocr_tool_demo.py	Demo of implementing an OCR tool
examples/09_remove_image_background_tool_demo.py	Demo of automatically removing image backgrounds, including automatic pip installation and calling the library to remove image backgrounds
examples/10_vision_demo.py	Vision understanding Demo
examples/11_web_search_openai_demo.py	Web search Demo based on OpenAI's function call
examples/12_web_search_moonshot_demo.py	Web search Demo based on Moonshot's function call
examples/13_storage_demo.py	Agent storage Demo
examples/14_custom_tool_demo.py	Demo of custom tools and autonomous selection and calling by large models
examples/15_crawl_webpage_demo.py	Demo of a webpage analysis workflow: crawling financing news from URLs, analyzing webpage content and format, extracting core information, and summarizing it into a markdown file
examples/16_get_top_papers_demo.py	Demo of parsing daily papers and saving them in JSON format
examples/17_find_paper_from_arxiv_demo.py	Demo of paper recommendation: automatically searching multiple groups of papers from arxiv, deduplicating similar papers, extracting core paper information, and saving it as a CSV file
examples/18_agent_input_is_list.py	Demo showing that the Agent's message can be a list
examples/19_naive_rag_demo.py	Basic RAG implementation, answering questions based on a text document
examples/20_advanced_rag_demo.py	Advanced RAG implementation, answering questions based on a PDF document, with new features: PDF file parsing, query rewriting, multi-path mixed recall (literal + semantic), and recall ranking (rerank)
examples/21_memorydb_rag_demo.py	Traditional RAG approach of placing reference materials in the prompt
examples/22_chat_pdf_app_demo.py	Demo of in-depth conversation with a PDF document
examples/23_python_agent_memory_demo.py	Demo of a Code Interpreter with memory, automatically generating and executing Python code, and retrieving results from memory on subsequent executions
examples/24_context_demo.py	Demo of conversation with context
examples/25_tools_with_context_demo.py	Demo of tools with context parameters
examples/26_complex_translate_demo.py	Demo of complex translation
examples/27_research_agent_demo.py	Demo of Research functionality, automatically calling search tools, summarizing information, and writing a scientific report
examples/28_rag_integrated_langchain_demo.py	RAG Demo integrated with LangChain
examples/29_rag_integrated_llamaindex_demo.py	RAG Demo integrated with LlamaIndex
examples/30_text_classification_demo.py	Demo of an Agent that automatically trains a classification model: reading the training set file and understanding the format, Google searching for the pytextclassifier library, crawling the GitHub page to understand how to call pytextclassifier, writing code and executing fasttext model training, and checking the prediction results of the trained model
examples/31_team_news_article_demo.py	Team implementation: team collaboration to write a news article, multi-role implementation, delegating different roles to complete their respective tasks: researcher retrieves and analyzes articles, writer writes the article according to the layout, and the results of multiple roles are summarized
examples/32_team_debate_demo.py	Team implementation: Demo of a two-person debate based on delegation, Trump and Biden debate
examples/33_self_evolving_agent_demo.py	Demo of a self-evolving Agent
examples/34_llm_os_demo.py	Initial design of LLM OS, based on LLM design operating system, can call RAG, code executor, Shell, etc. through LLM, and collaborate with code interpreter, research assistant, investment assistant, etc. to solve problems
examples/35_workflow_investment_demo.py	Workflow implementation for investment research: stock information collection, stock analysis, writing analysis reports, reviewing reports, and multiple tasks
examples/36_workflow_news_article_demo.py	Workflow implementation for writing news articles, multi-agent implementation, multiple calls to search tools, and generating advanced layout news articles
examples/37_workflow_write_novel_demo.py	Workflow implementation for writing novels: setting the novel outline, Google searching to reflect on the outline, writing the novel content, and saving it as a markdown file
examples/38_workflow_write_tutorial_demo.py	Workflow implementation for writing technical tutorials: setting the tutorial directory, reflecting on the directory content, writing the tutorial content, and saving it as a markdown file
examples/39_audio_multi_turn_demo.py	Demo of multi-turn audio conversation based on OpenAI's voice API
examples/40_web_search_zhipuai_demo.py	Demo of web search based on ZhipuAI's API, using the free glm-4-flash model and the free web-search-pro search tool

Self-evolving Agent

The self-evolving agent design:

Feature

Self-evolving Agents with Reflective and Memory-augmented Abilities (SAGE)

Implement:

Use PythonAgent as the SAGE agent and AzureOpenAIChat as the LLM, with code-interpreter functionality to execute Python code and automatically correct errors.
Use CsvMemoryDb as the memory for the SAGE agent to store user questions and answers, so that similar questions can be directly answered next time.

Run Self-evolving Agent App

cd examples
streamlit run 33_self_evolving_agent_demo.py

LLM OS

The LLM OS design:

Run the LLM OS App

cd examples
streamlit run 34_llm_os_demo.py

CLI

Support CLI for quick setup and experience of Agent:

code: cli.py

> agentica -h                                    
usage: cli.py [-h] [--query QUERY]
              [--model_provider {openai,azure,moonshot,zhipuai,deepseek,yi}]
              [--model_name MODEL_NAME] [--api_base API_BASE]
              [--api_key API_KEY] [--max_tokens MAX_TOKENS]
              [--temperature TEMPERATURE] [--verbose VERBOSE]
              [--tools [{search_serper,file_tool,shell_tool,yfinance_tool,web_search_pro,cogview,cogvideo,jina,wikipedia} ...]]

CLI for agentica

options:
  -h, --help            show this help message and exit
  --query QUERY         Question to ask the LLM
  --model_provider {openai,azure,moonshot,zhipuai,deepseek,yi}
                        LLM model provider
  --model_name MODEL_NAME
                        LLM model name to use, can be
                        gpt-4o/glm-4-flash/deepseek-chat/yi-lightning/...
  --api_base API_BASE   API base URL for the LLM
  --api_key API_KEY     API key for the LLM
  --max_tokens MAX_TOKENS
                        Maximum number of tokens for the LLM
  --temperature TEMPERATURE
                        Temperature for the LLM
  --verbose VERBOSE     enable verbose mode
  --tools [{search_serper,file_tool,shell_tool,yfinance_tool,web_search_pro,cogview,cogvideo,jina,wikipedia} ...]
                        Tools to enable

run：

pip install agentica -U
# single call, fill in the `--query` parameter
agentica --query "下一届奥运会在哪里举办" --model_provider zhipuai --model_name glm-4-flash --tools web_search_pro
# multi-turn conversation, do not fill in the `--query` parameter
agentica --model_provider zhipuai --model_name glm-4-flash --tools web_search_pro cogview --verbose 1

output:

2024-12-30 21:59:15,000 - agentica - INFO - Agentica CLI
>>> 帮我画个大象在月球上的图

> 我帮你画了一张大象在月球上的图，它看起来既滑稽又可爱。大象穿着宇航服，站在月球表面，背景是广阔的星空和地球。这张图色彩明亮，细节丰富，具有卡通风格。你可以点击下面的链接查看和下载这张图片：

![大象在月球上的图](https://aigc-files.bigmodel.cn/api/cogview/20241230215915cfa22f46425e4cb0_0.png)

Contact

Issue (suggestions) ：
Email me: xuming: xuming624@qq.com
WeChat me: Add my WeChat ID: xuming624, note: Name-Company-NLP to join the NLP group.

Citation

If you use agentica in your research, please cite it as follows:

APA:

Xu, M. agentica: A Human-Centric Framework for Large Language Model Agent Workflows (Version 0.0.2) [Computer software]. https://github.com/shibing624/agentica

BibTeX:

@misc{Xu_agentica,
  title={agentica: A Human-Centric Framework for Large Language Model Agent Workflows},
  author={Xu Ming},
  year={2024},
  howpublished={\url{https://github.com/shibing624/agentica}},
}

License

The license is The Apache License 2.0, free for commercial use. Please include a link to agentica and the license in the product description.

Contribute

The project code is still rough, if you have any improvements to the code, you are welcome to submit them back to this project. Before submitting, please note the following two points:

Add corresponding unit tests in tests
Use python -m pytest to run all unit tests and ensure all tests pass

Then you can submit a PR.

Acknowledgements

https://github.com/langchain-ai/langchain
https://github.com/simonmesmith/agentflow
https://github.com/phidatahq/phidata

Thanks for their great work!

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README_EN.md

README_EN.md

Agentica: Build AI Agents

📖 Introduction

Agent Components

Agentica Workflow

🔥 News

😊 Features

Install

Getting Started

1. Install requirements

2. Run the example

Web UI

Examples

Self-evolving Agent

Feature

Run Self-evolving Agent App

LLM OS

Run the LLM OS App

CLI

Contact

Citation

License

Contribute

Acknowledgements

Files

README_EN.md

Latest commit

History

README_EN.md

File metadata and controls

Agentica: Build AI Agents

📖 Introduction

Agent Components

Agentica Workflow

🔥 News

😊 Features

Install

Getting Started

1. Install requirements

2. Run the example

Web UI

Examples

Self-evolving Agent

Feature

Run Self-evolving Agent App

LLM OS

Run the LLM OS App

CLI

Contact

Citation

License

Contribute

Acknowledgements