Open Source Models
While OpenAI is generally recommended, there are situations where you might prefer open-source models. The following projects offer alternatives by mimicking the Assistants API:
✅ Tested Projects
- Astra Assistants API - The best and the easiest option for running Open Source models. Supports Assistants API V2. See example notebook.
- Open Assistant API - Fully local, stable and tested, but only supports Assistants V1. See example here
🔜 Other Projects
- OpenOpenAI - Unverified.
- LiteLLM - Assistants API Proxy in development.
Astra Assistants API
To use agency-swarm with Astra Assistants API, follow these steps:
1. Create an account on Astra Assistants API and obtain an API key.
2. Add Astra DB Token to your .env file:
Copy token from the file that starts with "AstraCS:" and paste it into your .env file.
3. Add other model provider API keys to .env as well:
PERPLEXITYAI_API_KEY=your_perplexityai_api_key
ANTHROPIC_API_KEY=your_anthropic_api_key
TOGETHER_API_KEY=your_together_api_key
GROQ_API_KEY=your_groq_api_key
4. Install the Astra Assistants API and gradio:
5. Patch the OpenAI client:
from openai import OpenAI
from astra_assistants import patch
from agency_swarm import set_openai_client
from dotenv import load_dotenv
load_dotenv()
client = patch(OpenAI())
set_openai_client(client)
6. Create an agent:
Create an agent and replace the model parameter with the name of the model you want to use. With Astra Assistants you can upload files like usual using files_folder
.
from agency_swarm import Agent
ceo = Agent(name="ceo",
description="I am the CEO",
model='ollama/llama3',
# model = 'perplexity/llama-3-8b-instruct'
# model = 'anthropic/claude-3-5-sonnet-20240620'
# model = 'groq/mixtral-8x7b-32768'
# model="gpt-4o",
files_folder="path/to/your/files"
)
7. Create an agency:
You can add more agents as needed, just make sure all manager agents support function calling.
8. Start gradio:
To utilize your agency in gradio, apply a specific non-streaming demo_gradio
method from the agency-swarm-lab repository:
from agency_swarm import Agency
from .demo_gradio import demo_gradio
agency = Agency([ceo])
demo_gradio(agency)
For a complete example, see the notebook.
General Instructions
To use agency-swarm with any other projects that mimic the Assistants API, generally, you need to follow these steps:
1. Install the previous version of agency-swarm as most projects are not yet compatible with streaming and Assistants V2:
2. Switch out the OpenAI client:
import openai
from agency_swarm import set_openai_client
client = openai.OpenAI(api_key="whatever", base_url="http://127.0.0.1:8000/")
set_openai_client(client)
3. Set the model parameter:
from agency_swarm import Agent
ceo = Agent(name="ceo", description="I am the CEO", model='ollama/llama3')
4. Start Gradio:
To utilize your agency in gradio, apply a specific non-streaming demo_gradio
method from the agency-swarm-lab repository:
from agency_swarm import Agency
from .demo_gradio import demo_gradio
agency = Agency([ceo])
demo_gradio(agency)
5. For backend integrations, simply use:
Limitations
- Function calling is not supported by most open-source models: This limitation prevents the agent from communicating with other agents in the agency. So, it must be positioned at the end of the agency chart and cannot utilize any tools.
- RAG is typically limited: Most open-source assistants API implementations have restricted Retrieval-Augmented Generation capabilities. It is recommended to develop a custom tool with your own vector database.
- CodeInterpreter is not supported: The Code Interpreter feature is still under development for all open-source assistants API implementations.
Future Plans
Updates will be provided as new open-source assistant API implementations stabilize.
If you successfully integrate other projects with agency-swarm, please share your experience through an issue or pull request.