Mullai v0.0.1

Mullai is an advanced AI Assistant framework built entirely on .NET. It provides a robust foundation for creating intelligent, multi-turn conversational AI agents equipped with 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.

Quick Install

Install the Mullai CLI directly to your system using our automated install scripts.

macOS / Linux (Bash)

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

Windows (PowerShell)

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

Key Features

  • Multi-Provider with Fallback: Integrate Gemini, Groq, Cerebras, Mistral, and more. If one fails, the next in line is automatically tried.
  • Centralized Model Catalog: Manage all model metadata (priority, pricing, context window) from a single models.json file.
  • Rich Tool Ecosystem: Empower agents with WeatherTool, CliTool, and FileSystemTool.
  • Robust Middleware: Pipeline for FunctionCalling, PII Handling, and Guardrails.
  • Memory & Skills: Persistent UserInfoMemory and dynamic AgentSkills.

Project Architecture

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

Core Layers

  • Abstractions: Fundamental interfaces and base classes.
  • Agents: The brain of Mullai, housing agent personalities and factory logic.
  • Channels: Interaction via CLI, API, Telegram, or WebAssembly.
  • Providers: Adapters for various LLM backends (Gemini, Groq, etc.).
  • Telemetry: Shared OpenTelemetry configuration for tracing and metrics.

Provider Configuration

Mullai separates model configuration from sensitive API keys for flexibility and security.

models.json — Model Catalog

Defined in src/Mullai.Global.ServiceConfiguration/models.json.

{
  "MullaiProviders": {
    "Providers": [{
      "Name": "Gemini",
      "Priority": 1,
      "Enabled": true,
      "Models": [{ "ModelId": "gemini-2.0-flash", ... }]
    }]
  }
}

appsettings.json — API Keys

Sensitive keys are stored in src/Mullai.Global.ServiceConfiguration/appsettings.json. 

{
  "Gemini": { "ApiKey": "YOUR_API_KEY" },
  "Groq": { "ApiKey": "YOUR_API_KEY" }
}

Observability

Deep insights out-of-the-box. Mullai integrates with OpenTelemetry to emit:

  • Distributed Traces: Parent spans for requests, child spans for each LLM attempt.
  • Structured Logs: Granular insight into the agent's decision-making process.
  • Metrics: Quantitative views of health and performance.

A local stack (Jaeger, Prometheus) can be launched via docker/observability.

Contributing

We welcome contributions! Whether you're adding providers, creating tools, or improving documentation.

  • Report Bugs: Use Github issues for bug reports or feature suggestions.
  • Development: Follow .NET best practices and our branching strategy.
  • Pull Requests: Ensure your code is well-tested and documented.

See our Full Contributing Guidelines for more details.