Interactive UI for Swarms Framework

[harshalmore31]

Interactive UI for Swarms Framework: Swarm-Specific Parsing, Agent Management, and LLM Integration with Hugging Face Deployment & Local Deployment

Description:

• Developed a robust and feature-rich UI for the Swarms framework using Gradio, focused on enhancing usability, modularity, and interactive capabilities. Key additions include:
  • Swarm-Specific Parsing:
    • Implemented parsers for all swarm types, including SequentialWorkflow, ConcurrentWorkflow, AgentRearrange, MixtureOfAgents, SpreadSheetSwarm, and auto.
    • Ensures consistent formatting of output data and detailed feedback for each workflow type.
  • Dynamic Prompt Loading:
    • Integrated functionality to load agent prompts dynamically from a JSON file.
    • Default prompts are used if the file is absent or improperly formatted.
  • Agent Management in UI:
    • Enabled creation of new agent prompts directly via the UI.
    • Supports editing and validation of custom prompts with seamless integration into the workflow configuration.
  • UI Framework Enhancements:
    • Introduced a dedicated UI class for streamlined Gradio component management, improving scalability and readability of the codebase.
    • Added modular components like dropdowns, sliders, and conditional inputs for an intuitive user experience.
  • LLM Integration:
    • Using LiteLLM, integrated the following models and providers:
      • OpenAI
      • Anthropic
      • Cohere
      • Gemini
      • Mistral
      • Groq
      • Perplexity

Key Features:

• Interactive Swarm Configuration: Supports flexible configuration for workflows, including custom execution flows for AgentRearrange and advanced features like agent flow validation.
• Agent-Centric Design: Facilitates agent selection, prompt management, and dynamic updates for added flexibility.
• User Feedback and Debugging: Incorporates detailed logging, error handling, and status displays for enhanced user experience.

Issue:

• Not linked to any specific issue. This PR focuses on establishing a functional and scalable UI layer for the Swarms framework.

Dependencies:

• Gradio: Core framework for building the modular UI.
• dotenv: Securely manages API keys and configurations.
• litellm: Enables seamless integration with multiple language model providers and their respective models.

Deployment:

• Hosted on Hugging Face Spaces. (https://huggingface.co/spaces/harshalmore31/Swarms)

Tag Maintainer:

@kyegomez

Twitter Handle:

@harshalmore31

---

Additional Changes:

• spreadsheet_swarm.py is modified accordingly for the working of UI

---

---

📚 Documentation preview 📚: https://swarms--727.org.readthedocs.build/en/727/

[evelynmitchell]

Thanks Harshal!

In the HuggingFace space, the contrast in the UI is very low (grey on grey).

There is also no visual indicator that 'Select Provider' is a pulldow menu, and the text in it is even lighter than in the text box.

Can you increase the contrast a bit?

As a feature improvement, tags to filter the Agents on would be helpful (role, business department, frequently used with, last updated, ...) A long pulldown menu gets unusable quickly.