AgentCraftLab: A .NET-Native AI Agent Platform

AgentCraftLab is an open-source .NET-native AI agent platform that supports visual workflow design, multi-protocol deployment, autonomous agent execution, and a built-in search engine. It allows building complete AI agent capabilities within the .NET ecosystem without Docker or external databases, filling the gap of AI agent solutions in the .NET ecosystem.

With the development of AI agent technology, platforms like Flowise and Dify are mostly based on Python/Node.js or rely on Docker, requiring .NET teams to maintain an additional tech stack. AgentCraftLab was created to address this issue—it is fully built on .NET, allowing .NET developers to design, test, and deploy agent workflows without leaving their familiar ecosystem.

The platform adheres to the principles of .NET-native, zero external dependencies, and local-first. It uses SQLite by default without needing an external database, does not mandate Docker (can be directly deployed on the .NET runtime), and provides a visual workflow editor. This architecture is easy to deploy, fast to start, and has low resource consumption—ideal for local/private environments. It is the only open-source solution that simultaneously meets the requirements of .NET-native, no Docker, built-in Teams bot, and local SQLite.

• Visual Workflow Studio: A drag-and-drop editor based on React Flow, supporting over 10 node types such as Agent, Condition, Loop, Parallel, and Human-in-the-Loop, enabling the construction of complex workflows. The Human-in-the-Loop node allows pausing during execution to wait for user input/approval.
• AI Construction Mode: Users describe their goals in natural language, and AI automatically generates workflows, lowering the barrier to entry and facilitating rapid prototyping.

• Multi-Protocol Deployment: Supports A2A (Agent-to-Agent communication), MCP (integration with Claude/ChatGPT), REST API, and Teams Bot endpoints. Design once and deploy to multiple places—all endpoints are authenticated via API keys.
• Advanced Execution Modes: Autonomous Agent (ReAct loop architecture for independent decision-making steps); Flow Mode (AI plans a sequence of nodes, which crystallizes into a reusable workflow after execution).

• Built-in CraftSearch: Combines full-text search, vector search, and RRF hybrid ranking, supporting content extraction from multiple formats like PDF/DOCX.
• RAG Pipeline: Automatically extracts, chunks, embeds, and indexes uploaded documents to build temporary/persistent knowledge bases.
• Doc Refinery: Extracts structured data from unstructured documents, offering fast (single-round LLM) and precise (multi-layer agent + verification) modes.

AgentCraftLab fills the gap of AI agent platforms in the .NET ecosystem, offering features comparable to mainstream platforms. With its .NET-native, zero external dependency, and local-first architecture, it provides an ideal choice for .NET teams and is worth the attention and contribution of .NET organizations.