Heym: Open-Source Self-Hosted AI-Native Workflow Automation Platform

Heym is an AI-native automation platform designed from the ground up for LLMs and agents, offering a visual canvas, multi-agent orchestration, RAG pipelines, human-in-the-loop collaboration, and MCP support. It can compete with commercial platforms like n8n and Zapier. As a fully self-hosted open-source platform, it marks a shift in the automation paradigm—AI is no longer an add-on feature but the execution model itself, making it suitable for users and teams that value data sovereignty and require deep customization.

The workflow automation space has long been dominated by tools like Zapier and Make.com, which are based on the classic 'trigger-action' model and add AI capabilities as an afterthought. As large language models become the core of computing, this architecture is gradually becoming inadequate. The emergence of Heym brings about a paradigm shift: AI is not an add-on but the execution model itself. It is a fully self-hosted open-source platform that integrates a visual workflow editor, AI agents, RAG, human-in-the-loop (HITL) collaboration, and MCP into a single runtime, designed from the start for LLMs and agents.

The core of Heym is a visual canvas based on Vue Flow, supporting over 30 node types. Users can drag and drop to build complex workflows without code. The platform offers first-class LLM nodes and full Agent nodes: Agent nodes support tool calling, Python tools, MCP connections, a skill system, and optional persistent memory (extracted in the background via knowledge graphs); the LLM Batch API mode provides real-time status branching for supported providers, allowing monitoring of batch task progress. Multi-agent orchestration is a key feature: a main agent coordinates multiple named sub-agents and sub-workflows, with relationships visualized on the canvas, making the construction of complex layered AI systems more intuitive.

Human-In-The-Loop (HITL) Collaboration

Fully automated AI systems require human oversight at critical decision points. Heym's HITL allows agents to pause during execution, request user approval/input before continuing, and capture a snapshot of the entire execution state. Users can view the full context during review, which is particularly important in safety-critical scenarios (e.g., financial transfers).

Built-In RAG and Vector Storage

Heym provides out-of-the-box RAG capabilities. Users can insert documents and run semantic searches in dedicated nodes, with QDrant vector storage used under the hood. The platform manages embedding and sharding details, lowering the entry barrier for RAG while also supporting integration with external vector stores.

MCP (Model Context Protocol) is an open standard launched by Anthropic that allows AI models to safely access external tools and data sources. Heym fully supports MCP: Agent nodes can act as MCP clients to connect to any MCP server, and users can also expose workflows as MCP servers for clients like Claude and Cursor to call. This bidirectional support makes Heym a hub in the AI ecosystem—leveraging community MCP tools while allowing users' workflows to be seamlessly called by other AI applications, which is a key difference from closed commercial platforms.

Auto-Repair and Fault Tolerance

Web automation often faces selector failure issues. Heym's Auto Heal feature can automatically detect and fix Playwright selector failures; the LLM Fallback mechanism automatically switches to a backup model when the main model fails, ensuring workflow continuity; users can also configure inference strength and temperature for each Agent node to finely control AI behavior.

Observability and Evaluation

Heym provides complete LLM tracing functionality, recording requests, responses, tool calls, and time consumption for each agent invocation to aid debugging and optimization; the built-in evaluation (Evals) system allows users to define test suites and run evaluations with one click, quantifying the impact of workflow modifications on performance.

Heym competes directly with commercial platforms like n8n, Zapier, and Make.com. Its advantages include: native LLM batch API support, per-agent persistent memory, automatic context compression, built-in WebSocket support, natural language workflow construction, skill system, Auto Heal, and full open-source self-hosting. Commercial platforms still have advantages in ecosystem maturity, number of pre-built integrations, and enterprise support. Heym is more suitable for users and teams with strong technical capabilities, who value data sovereignty and require deep customization.

Heym supports one-click deployment via Docker, allowing users to run the full platform on their own infrastructure; the project uses the CC+MIT license, permitting free use and modification; the product website heym.run provides an online experience and detailed documentation. For users looking to explore the new paradigm of AI workflow automation, Heym is a feature-rich, modern-architecture open option—it is not just a tool, but an exploration of the future form of AI-native automation platforms.