Luann (fka TypeAgent) allows you to create a LLM agent,which has complete memory module (long-term memory, short-term memory) and knowledge module（Various knowledge bases）,eventually become an agent with both EQ and IQ.Another most important thing is that he can use various tools and become a real tool man AND Open Source.IT can also create various types of agents,IT also called LLM OS.
luann python client and luann docker will coming soon.....
Luann makes it easy to build and deploy stateful LLM agents with support for:
- Long term memory/state managemet,short term memory,knowledge base long term memory
- Basic RAG workflow for knowledge base which created by external data sources (e.g. PDF files)
- Intelligent switching knowledge base
- Defining and calling custom tools，then help you do something(take action)
- Changing personas settings and role-playing
- Complete evauation subsystem for rag and whole agent system at Production Environment
- Reasoning system(from phidata agent)
- ADE UI
- Create various types of agent(swe,ai scientist,etc..)
You can also use Luann to deploy agents as a service. You can use a Luann server to run a multi-user, multi-agent application on top of supported LLM providers.
IF you have any questions about using Luann ,you can make a issues
IF you clone this repo or are interesting ,please click star ,which will make more people to see
- Clone the repo
- Run
python main.py configure - Run
python main.py server - Go to
localhost:8283in the browser to view the developer portal
Once the server is running, you can use the [REST API] to connect to localhost:8283 (if you're running with the CLI) to create users, agents, and more.
coming......
Luann is designed to Create various types of agent. The following type of agent are supported:
| Type | supported |
|---|---|
| Memgpt | ✅ |
| openhands | ❌ |
Luann is designed to Create various types of agent.
Luann is designed to be model and provider agnostic. The following LLM and embedding endpoints are supported:
| Provider | LLM Endpoint | Embedding Endpoint |
|---|---|---|
| OpenAI | ✅ | ✅ |
| Azure OpenAI | ✅ | ✅ |
| Google AI (Gemini) | ✅ | ❌ |
| Anthropic (Claude) | ✅ | ❌ |
| Groq | ✅ (alpha release) | ❌ |
| Cohere API | ✅ | ❌ |
| vLLM | ✅ | ❌ |
| Ollama | ✅ | ✅ |
| LM Studio | ✅ | ❌ |
| koboldcpp | ✅ | ❌ |
| oobabooga web UI | ✅ | ❌ |
| llama.cpp | ✅ | ❌ |
| HuggingFace TEI | ❌ | ✅ |
| baichuan LLM | ❌ | ❌ |
| qianwen LLM | ❌ | ❌ |
When using Luann with open LLMs (such as those downloaded from HuggingFace), the performance of Luann will be highly dependent on the LLM's function calling ability ,Language understanding and reasoning skills.
As you can see, typeagent will intelligently change the memory
- Add Luann client
- Add other type agent( more complex agent like openhand, or simple functional agent)
- test vectordb and other llms
- add baichuan/qianwen etc LLM
- add voice clone （tortoise-tts）
- add Agent Evaluation
- add Production RAG Evaluation（Ragas https://github.com/explodinggradients/ragas and part from crewai）（progess 70%）
- add openai swarm
- add LLM reranker subsystem（GPT4o reranker）
- add automation workflow（not only chat but also do something）
- add session timeline （from phidata）
- add Other modulus models, such as OCR models, BLIP etc.
- add trained SLM or SMM or fine-tune SLM by using Self-developed deep learning training framework PyFlame ,which aimed to use specific usecase,because Special small models act as lubricants for modules in the agent framework
- add usage recording
- This project is a Leisure time hobby，If you are interested in the project ,you can make a issue.
- Our codebase for the Luann builds heavily on MemGPT codebase,Thanks for open-sourcing!
- The difference of MemGPT(letta) and Luann is that Luann optimizes the entire memgpt code structure and creatively adds a complete memory module and knowledge base module and add SML.
- The other main difference of MemGPT(letta) and Luann is that Luann can create various type of agent,one architecture ,one pipline,one framework.
- Documentation temporary reference MemGPT codebase.Thanks letta
- New ideas and new features will be added continuously,make everyone use very well
- new version is coding ,please wait..
goal: EQ and IQ AGENT or a tool man(not only chat but also do something automately) ,or briage of llm and application,finally to acting agent