Workflow Editor

Core editor capabilities

• Drag/drop node palette
• Node-to-node edge connections
• Curved edges
• Node configuration via modal
• Node delete (Delete key)
• Save/import/export workflows
• Execution logs pane and run-inputs pane
• Per-node execution status visualization
• Agent port + buttons that open filtered drawers for Language Models, Memory, Tools, and Workers

Typical flow patterns

1. Basic LLM:
   • text_input -> prompt_template -> ollama_chat_model -> output
2. Agentic tool loop:
   • webhook_input -> agent_orchestrator -> output
   • Attachments:
     • chat_model: any Chat Model node, including OpenAI, Anthropic, Ollama, OpenAI Compatible, AI Gateway, Azure OpenAI, Gemini, or legacy llm_call
     • memory: local_memory
     • tool: one or more mcp_tool
3. RAG:
   • text_input -> rag_retrieve -> prompt_template -> llm_call -> output

Validation behavior

Validation checks include:

• Graph structure and required edges
• Required node config fields
• Attachment-type correctness (for agent/supervisor ports)
• Node-type specific rules

Non-connected nodes are not executed.