Mullai — AI Agents & Multi-Agent Workflows

Build Intelligent, Extensible, and Observant AI Agents and Multi-Agent workflows with .NET

---

Note: Active Research & Development Mullai is currently in active research and development. While we strive for stability, you may encounter bugs, incomplete features, or breaking changes. We appreciate your patience and welcome contributions to help improve the project!

---

Mullai is a powerful and flexible AI Agent platform built entirely on .NET. It provides a robust foundation for creating intelligent, multi-turn conversational AI agents and multi-agent workflows that are equipped with a rich set of tools, memory, and skills. Leveraging Microsoft.Extensions.AI and Microsoft.Agents.AI, Mullai empowers developers to build sophisticated AI assistants with a highly scalable and observable architecture.

🚀 Unified Workflow Dashboard (Web UI)

The primary interface for Mullai is now a modern Web Dashboard built with Blazor Server. It provides a premium environment for orchestrating multi-agent systems and monitoring complex workflows.

• Agent Orchestrator: Build, run, and monitor multi-agent workflows in real-time.
• Interactive Chat: Fluid conversations with rich tool-call visualization and streaming.
• Visual Task Graph: See the dependency tree of your agents as they collaborate.
• Observability Hub: Monitor provider performance, token usage, and execution traces.
• Centralized Settings: Effortlessly manage LLM providers and model configurations in one place.

---

🏗️ Multi-Agent Workflows

Mullai features a robust Workflow Engine designed for complex orchestration beyond simple chat.

• Dynamic Execution: Support for Single-Agent and Parallel-Agent execution strategies.
• Webhook Integration: Trigger workflows via standard webhooks for seamless automation.
• Execution History: Full persistence and auditing of every task run with detailed state snapshots.

---

💻 Developer CLI

For terminal-focused development and quick testing, Mullai continues to offer a Rich Console Application (powered by Spectre.Console). It provides high-performance interaction, real-time tool-call insights, and a multi-panel layout for deep debugging directly in your shell.

---

Quick Install (Binary)

macOS / Linux (Bash)

Run the following in your terminal:

curl -sSL https://raw.githubusercontent.com/agentmatters/mullai-bot/main/scripts/install.sh | bash

Windows (PowerShell)

Run the following in PowerShell (as Administrator if you want to update System PATH, though User PATH is default):

irm https://raw.githubusercontent.com/agentmatters/mullai-bot/main/scripts/install.ps1 | iex

This will:

1. Detect your Operating System and Architecture.
2. Download and extract the latest Mullai package.
3. Install the CLI to ~/.mullai/bin (or $HOME\.mullai\bin on Windows).
4. Install the Web app to ~/.mullai/web and register it as a system service (listening on port 7755).
5. Add the CLI installation directory to your PATH.

---

Uninstall

macOS / Linux (Bash)

curl -sSL https://raw.githubusercontent.com/agentmatters/mullai-bot/main/scripts/uninstall.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/agentmatters/mullai-bot/main/scripts/uninstall.ps1 | iex

This will stop and remove the Mullai Web service and delete the installation directory.

---

Service Management

After installation, the Mullai Web service runs automatically. You can manage it using the following commands:

macOS / Linux

• Start:
  • macOS: launchctl load ~/Library/LaunchAgents/com.mullai.bot.web.plist
  • Linux: sudo systemctl start mullai-web
• Stop:
  • macOS: launchctl unload ~/Library/LaunchAgents/com.mullai.bot.web.plist
  • Linux: sudo systemctl stop mullai-web
• Restart:
  • macOS: launchctl unload ~/Library/LaunchAgents/com.mullai.bot.web.plist && launchctl load ~/Library/LaunchAgents/com.mullai.bot.web.plist
  • Linux: sudo systemctl restart mullai-web
• Remove Service:
  • macOS:

    launchctl unload ~/Library/LaunchAgents/com.mullai.bot.web.plist
    rm ~/Library/LaunchAgents/com.mullai.bot.web.plist

  • Linux (systemd):

    sudo systemctl stop mullai-web
    sudo systemctl disable mullai-web
    sudo rm /etc/systemd/system/mullai-web.service
    sudo systemctl daemon-reload

  • Linux (user service):

    systemctl --user stop mullai-web
    systemctl --user disable mullai-web
    rm ~/.config/systemd/user/mullai-web.service
    systemctl --user daemon-reload

Windows (PowerShell)

• Start: Start-Service MullaiWeb
• Stop: Stop-Service MullaiWeb
• Restart: Restart-Service MullaiWeb

---

Key Features

Mullai is designed for resilience, flexibility, and developer-friendliness:

• Multi-Agent Workflow Engine: Define complex collaboration patterns (sequential or parallel) with full execution history, state persistence, and webhook support.
• Multi-Provider with Automatic Fallback: Seamlessly integrates Gemini, Groq, Cerebras, Mistral, OpenRouter, and Ollama. If one provider fails, Mullai automatically fails over to the next available model.
• Unified Web Dashboard: A premium Blazor-based interface for interactive chat, workflow orchestration, and real-time visualization of agent task graphs.
• Rich Tool Ecosystem: Empower agents with a diverse set of capabilities ranging from system automation to document processing.
• Observability Built-in: Full OpenTelemetry integration providing deep insights via distributed traces, structured logs, and performance metrics.
• Robust Middleware Pipeline: Intercept and process messages at any stage with built-in support for tool-calling, PII scrubbing, and safety guardrails.
• Persistent Memory & Skills: Advanced state management with recursive task trees and dynamic, file-based agent skills.

Supported Tools

Mullai agents come equipped with a versatile suite of tools to interact with the environment and external services:

Category	Tool	Description
Orchestration	WorkflowTool	Manage, upsert, and execute complex multi-agent workflows.
Automation	BashTool	Execute shell commands and scripts in a terminal environment.
	CliTool	Run command-line interface tools and capture output.
Development	CodeSearchTool	Perform advanced search and navigation within local codebases.
	FileSystemTool	Read, write, list, and manage files and directories.
Web & Data	WebTool	Search the web and fetch/summarize content from any URL.
	RestApiTool	Make authenticated HTTP requests (GET, POST, etc.) to external APIs.
	HtmlToMarkdown	Convert raw HTML content into clean, readable Markdown.
Documents	WordTool	Create and edit Microsoft Word (.docx) documents via OpenXML.
Productivity	TodoTool	Manage project-specific task lists and todo items.
	WeatherTool	Retrieve real-time weather data and geolocation information.

---

Project Architecture

Mullai is built with a modular and decoupled architecture, promoting maintainability and extensibility.

graph TD;
    Agents[Mullai.Agents] --> Abstractions[Mullai.Abstractions];
    Agents --> Tools[Mullai.Tools];
    Agents --> Memory[Mullai.Memory];
    Agents --> Skills[Mullai.Skills];
    Agents --> Middleware[Mullai.Middleware];
    
    Web[Mullai.Web] --> Agents;
    Web --> Workflows[Mullai.Workflows];
    Web --> Providers[Mullai.Providers];
    Web --> Config[Mullai.Global.ServiceConfiguration];
    Web --> Telemetry[Mullai.Telemetry];
    Web --> Logging[Mullai.Logging];

    Workflows --> TaskRuntime[Mullai.TaskRuntime];
    Workflows --> Agents;
    
    CLI[Mullai.CLI] --> Agents;
    CLI --> Providers;
    CLI --> Config;
    CLI --> Telemetry;
    CLI --> Logging;

    Providers --> MullaiChatClient[MullaiChatClient];
    MullaiChatClient --> Gemini[Gemini];
    MullaiChatClient --> Groq[Groq];
    MullaiChatClient --> Cerebras[Cerebras];
    MullaiChatClient --> Mistral[Mistral];
    MullaiChatClient --> OpenRouter[OpenRouter];
    MullaiChatClient --> Ollama[Ollama];
    MullaiChatClient --> More[...];

    ChannelsCore[Mullai.Channels.Core] --> Agents;
    ChannelsTelegram[Mullai.Channels.Telegram] --> ChannelsCore;
    ChannelsApi[Mullai.Channels.Api] --> ChannelsCore;

Loading

Core Components Explained:

• Mullai.Abstractions: Defines core interfaces and base classes, ensuring a consistent and extensible foundation.
• Mullai.Agents: The brain of Mullai, housing the AgentFactory and definitions for various AI agent personalities.
• Mullai.Channels.*: Projects like Mullai.Channels.Core, Mullai.Channels.Api, and Mullai.Channels.Telegram enable interaction with agents via different communication platforms (e.g., API, Telegram, WebAssembly).
• src/Mullai.CLI: Provides a Rich Console Application for interactive agent conversations directly in the terminal.
• Mullai.Global.ServiceConfiguration: Centralizes application settings, including LLM model priorities and API keys.
• Mullai.Logging: Provides mechanisms for comprehensive logging, including LLM request/response details.
• Mullai.Memory & Mullai.Skills: Manages conversational state, user-specific data, and dynamic, reusable agent capabilities.
• Mullai.Middleware: Offers a powerful interception layer for applying cross-cutting concerns like security, data privacy, and function calling.
• Mullai.Providers: Integrates with various LLM backends, abstracting away their differences to provide a unified MullaiChatClient.
• Mullai.Telemetry: Implements shared OpenTelemetry configuration for distributed tracing, metrics, and structured logging across the entire framework.
• Mullai.Tools: Contains external capabilities that agents can invoke, allowing them to interact with the operating system, file system, and other services (e.g., weather data).

---

🌐 Supported Providers

Mullai is designed to be provider-agnostic, supporting a wide range of LLM backends through both native integrations and OpenAI-compatible endpoints:

• Google Gemini: Native support for Gemini 1.5 Pro, Flash, and 2.0.
• Groq: Ultra-fast inference for Llama 3.1, Mixtral, and Gemma.
• Cerebras: High-performance compute for Llama 3.1 models.
• Mistral AI: Native support for Mistral Large, Pixtral, and Codestral.
• NVIDIA NIM: Optimized inference through NVIDIA's model catalog.
• OpenAI: Support for GPT-4o, GPT-4 Turbo, and GPT-4o mini.
• OpenRouter: Unified access to over 200+ models including DeepSeek, Claude 3.5, and Llama 3.1.
• Ollama: Local execution of various open-source models via OpenAI-compatible API.

Getting Started

Follow these steps to get Mullai up and running quickly.

Prerequisites

• (Required for Building) .NET 10 SDK
• (Optional) Docker Desktop for running the OpenTelemetry observability stack (Jaeger, Prometheus).
• An API key for at least one supported LLM provider (e.g., Gemini, Groq, Cerebras, Mistral, OpenRouter) or a local Ollama instance configured to serve an LLM.

Setup and Run (From Source)

1. Clone the repository:

   git clone https://github.com/agentmatters/mullai-bot.git
   cd Mullai

2. Run the Web Dashboard (Primary):

   cd src/Mullai.Web
   dotnet run

   Access the UI at http://localhost:7755 (or as specified in the console).

3. (Optional) Run the Developer CLI:

   cd src/Mullai.CLI
   dotnet run

Contributing

We welcome contributions from the community! Whether you want to add new LLM providers, create innovative tools or middleware, improve the existing Blazor UI, enhance the CLI, or improve documentation, your efforts are highly appreciated.

Please review our Contributing Guidelines for detailed information on:

• Code of Conduct
• How to report bugs or suggest enhancements
• Development best practices for C# and .NET
• GitHub workflow, including branching strategy and commit message format
• Adding new features like LLM providers, tools, middleware, and agent personalities
• Security considerations and licensing

---

License

This project is licensed under the MIT License. See the LICENSE file for details.