Potery Multi-Agents System

multi-agent system for creating poetry with accompanying text-to-speech and music.

Multi-Agent System: Poetry, Speech, and Music

Generated Potery Example

POEM: 八月7日凌晨

七月初七鹊桥边，
银汉迢迢星光寒。
人间几度离合苦，
梦里相逢泪湿衫。
晓风残月催人醒，
又是人间一日忙。
愿得此情长久在，
不负韶华不负卿。

final Poem:

This system will be composed of three distinct agents, each with a specialized function. They will work in a sequential workflow, passing the output of one agent as the input to the next.

Agent Roles:

Poet Agent: This agent's primary role is to generate a poem or limerick based on a theme or content provided by the user. It will focus on creativity, structure, and lyrical quality.

Voice Agent: This agent will take the text of the poem and convert it into speech. It will be responsible for creating a clear and expressive audio representation of the poem.

Music Agent: This agent will compose a short piece of music to serve as a background for the spoken poem. The goal is to create an atmosphere that complements the mood and theme of the poetry.

Workflow:

Input: The user provides a theme or some initial content to the PoetryAgent.

Poem Generation: The PoetAgent uses the input to create a poem.

Text-to-Speech: The generated poem is passed to the VoiceAgent, which converts the text into an audio file.

Music Composition: The MusicAgent creates a musical piece.

Final Output: The system will provide the poem's text, the spoken version of the poem, and the background music.

Text to Speech

bosonai/higgs-audio-v2-generation-3B-base

https://huggingface.co/bosonai/higgs-audio-v2-generation-3B-base

TextToSpeechTool: This tool now connects to the Gradio server you specified. It uses an async function to handle the connection and the tool call.

MCP Connection: The tool uses stdio_client and ClientSession from the mcp library to communicate with the server.

Task Dependency: I've added a context to the speak_poem_task to ensure it runs after the write_poem_task is complete.

Music Generation

facebook/musicgen-small

https://huggingface.co/facebook/musicgen-small

Start the ACP Agent

Start the main agent server:

uv run agent.py

Expected output:

🚀 Agent server starting on http://localhost:8000

Test Poetry Generation

Test the full poetry workflow via HTTP:

curl -X POST http://localhost:8000/runs \
  -H "Content-Type: application/json" \
  -d '{
        "agent_name": "poetry_agent",
        "input": [
          {
            "role": "user",
            "parts": [
              {
                "content": "春天的花朵",
                "content_type": "text/plain"
              }
            ]
          }
        ]
      }'