demo

Summary

This demo application is maintained as an experimental development environment and a showcase for library features. You are encouraged to use the source as a reference but beware that the quality is intended to be proof-of-concept only.

Example Classes

By default the demo includes the following two Agent classes:

• OpenAIFunctionAgent - An LLM agent that uses the OpenAI function calling API
• Host - An agent that exposes operating system access to the host system

More agent examples are located under the ./agents directory.

Running the demo

The demo application uses docker-compose for orchestration. Configuration is included for running the demo using the different space types. To run the demo:

1. Ensure you have Docker installed on your system.

2. Run:

    git clone git@github.com:operand/agency.git
    cd agency/examples/demo
    cp .env.example .env
   

3. Open and populate the .env file with appropriate values.

4. Start the application.

   To run the LocalSpace application:

   ./demo run local

   To run the AMQPSpace application:

   ./demo run amqp

5. Visit http://localhost:8080 and try it out!

The Gradio UI

The Gradio UI is a Chatbot based application used for development and demonstration.

It is defined in examples/demo/apps/gradio_app.py and simply needs to be imported and used like so:

from examples.demo.apps.gradio_app import GradioApp
...
demo = GradioApp(space).demo()
demo.launch()

The Gradio application automatically adds its user (you) to the space as an agent, allowing you to chat and invoke actions on the other agents.

The application is designed to convert plain text input into a say action which is broadcast to the other agents in the space. For example, simply writing:

Hello, world!

will invoke the say action on all other agents in the space, passing the content argument as Hello, world!. Any agents which implement a say action will receive and process this message.

Gradio App - Command Syntax

The Gradio application also supports a command syntax for more control over invoking actions on other agents.

For example, to send a point-to-point message to a specific agent, or to call actions other than say, you can use the following format:

/agent_id.action_name arg1:"value 1" arg2:"value 2"

A broadcast to all agents in the space is also supported using the * wildcard. For example, the following will broadcast the say action to all other agents, similar to how it would work without the slash syntax:

/*.say content:"Hello, world!"