What it does BotSharp is a .NET Core framework for building AI agent platforms inside enterprise codebases. It handles conversation state, multi-agent routing, RAG vector search, and plugs into a dozen LLM providers through a unified interface. A separate SvelteKit admin UI lets you build, test, and audit agents without writing front-end code.

The interesting bit The project treats C# as a feature, not a handicap: it implements ML algorithms directly in typed code and leans on .NET’s plugin pipeline to keep LLM providers, storage backends, and messaging channels (Telegram, WeChat, Slack, Twilio) fully decoupled. For teams already drowning in business logic written in C#, this means refactoring across “the system scope” stays sane.

Key highlights

• Supports GPT-4o/o1, Claude 3.5, Gemini 2, DeepSeek V3, LLaMA 3, and others via swappable plugins
• Built-in MCP integration with visual tool management for external service calls
• Multi-agent cooperation with state management and planning approaches for simple-to-complex tasks
• RAG interfaces with memory-based vector search (Qdrant plugin available)
• RESTful and WebSocket APIs if you want to skip the bundled SvelteKit UI
• Apache 2.0 licensed; runs on .NET 6+ across Windows, Linux, and macOS

Caveats

• Roadmap shows A2A, Computer Use, and Browser Use are still unchecked; some cutting-edge agent patterns are aspirational
• The “machine learning algorithms in C# directly” claim is stated but not benchmarked or detailed in the README
• Documentation exists on ReadTheDocs and a secondary Chinese site, which may mean quality or completeness varies

Verdict Worth a look if you’re a .NET enterprise team that needs to graft LLM agents onto existing systems without shipping Python microservices. If you’re already committed to Python’s agent ecosystem or need mature browser/computer-use automation today, this will feel early.