Bowen is an enterprise-grade platform for developing and orchestrating large model applications, enabling businesses to quickly, securely, and efficiently build, deploy, and manage LLM-based intelligent applications (such as intelligent assistants, automated workflows, and multi-agent collaboration systems).
Three application construction methods cater to different technical skill levels: ① Assistants – Designed for no-code business users, similar to OpenAI GPTs, using natural language prompts to quickly define roles and tasks, with support for binding knowledge bases, tools, and skills. ② Workflow – Aimed at low-code orchestration, leveraging the Langgraph engine for visual drag-and-drop node configuration, supporting human-in-the-loop interaction, multi-turn looped dialogues, rich node types, and version management. ③ Skill – Targeted at professional developers, based on deep component orchestration with LangChain/Langflow, featuring 18 major component categories (Agents, Chains, Loaders, Embeddings, LLMs, Memories, Prompts, Retrievers, Vector Stores, etc.), allowing highly customized complex AI pipelines.
Tools and Extension Ecosystem: The system includes built-in Bing/Google search, DALL-E 3 image generation, Python code executor, Tianyancha, economic and financial data interfaces, etc. It supports integration with enterprise-owned systems via the OpenAPI specification and proactively supports the MCP protocol, enabling access to external MCP servers through SSE/stdio.
Multimodal Knowledge Base Management: Document knowledge base (PDF/Word/Excel/PPT/images in all formats with intelligent chunking strategies), QA knowledge base (fixed question-answer pairs with AI-generated similar questions), and graph knowledge base (visual graph management supporting multi-hop reasoning).
Enterprise Database and NL2SQL: Supports MySQL/PostgreSQL, with custom field business semantic descriptions and pre-configured SQL examples to improve the accuracy of natural language to SQL conversion. Model Management and Security Compliance: Unified management of multiple models (OpenAI, Azure, Zhipu, Baidu Qianfan, Tongyi Qianwen, iFlytek Spark, MiniMax, Anthropic, and local private frameworks like Ollama, vLLM, Xinference, etc.). Content safety review (online/batch management of sensitive words, with automatic replacement of preset replies upon detection). Enterprise-grade permissions and auditing (user role management, session auditing, operation log auditing).
Typical Application Scenarios: Multi-turn intelligent customer service dialogues, automatic generation of professional reports (chapter splitting + template filling), intelligent document/contract review, NL2SQL natural language data queries, and multi-agent collaboration (AutoGen framework with roles such as Assistant, Coder, User, and GroupChatManager).
Technology Industrial Park Digital Platform A technology industrial park operator had massive digital deliverables, e.g. BIM/CAD/equipment ledgers, that were scattered and inefficient to retrieve. We delivered a unified digital and intelligent operating platform within 16 months.
Thoracic Surgery Triage & Follow-up System A Chinese public tertiary hospital reported that thoracic surgery appointment slots are often occupied by patients with non-adjuvant conditions, resulting in a misbooking rate of approximately 30%. Discharge follow-ups rely entirely on manual telephone calls, which is inefficient. Patient data must be deployed locally and cannot be transferred outside the local area network.