Nexent

Nexent is an MIT-licensed agent development platform from ModelEngine-Group that turns a plain-language description into a working multi-agent system, complete with tools, memory, and a knowledge base. It's built around 'Harness Engineering' principles and the Model Context Protocol (MCP), so agents are wired with constraints, feedback loops, and control planes rather than left as a free-firing LLM. Out of the box it ships with a two-tier memory system, an A2A (agent-to-agent) collaboration protocol, multi-modal I/O (voice, text, image, files), and ingestion for 20+ document formats.

Where Nexent differs from drag-and-drop competitors like Dify or Flowise is that there's no canvas: you describe the agent you want and it generates the orchestration for you. It's OpenAI-compatible against any provider and covers LLM, embedding, VLM, STT, and TTS slots, including Chinese domestic models, which makes it interesting for teams that need to swap backends. Self-hosted via Docker or Kubernetes, with multi-tenancy, RBAC, agent version management, and an agent marketplace aimed squarely at enterprise deployment.

It's open source with no managed-SaaS pricing page, so 'cost' is whatever your infra and model-API bills add up to. The project is active (5k+ stars, 200+ contributors, v2.0 shipped) but still young, and serious adopters should expect to operate it themselves.