Skip to main content

StormBPMN AI Assistant

About 1 min

StormBPMN AI Assistant

The AI Assistant is a chat built into the application that reads your diagrams, processes, roles, and documents, answers questions from the knowledge base, and actually creates and edits BPMN diagrams, assigns roles, and links documents — by invoking tools, rather than fabricating results.

All AI functionality is included in the AI module of the license (enableAiFeatures) and is enabled on the server via the environment variable MCP_ENABLED. Fine-tuning is available in the admin panel, under the “🤖 AI Assistant” section.

pgvector Required

The vector engine (which powers the knowledge base and the platform's fast semantic search) requires the pgvector extension in PostgreSQL. This is a basic platform requirement, not part of the AI module — see “PostgreSQL Requirements”.

Sections

  1. Enabling and Connecting a Process Model — how to enable the chat (ENV + license + toggle) and connect a language model: API format, base URL, token, separate write model, prompts.
  2. Knowledge Base (RAG) — uploading reference documents so that the chat responds based on their content; indexing and search parameters.
  3. Self-hosted LLM — how to run your own model on your GPU: model selection (benchmark) and deployment on vLLM.
  4. Voice Input — dictating messages in the chat: deploying the storm-media speech recognition service (Whisper) and settings in the admin panel. Available from version 6.6.6668.
  5. File Attachments and Diagram Import — files in chat messages: documents (pdf/docx/txt, etc.) are sent to the model context, while BPMN / draw.io / Visio diagrams are converted into Storm diagrams with a single click. Available in versions after 6.6.6955.
  6. Warm-ups (Storm Master) — personalized emails to employees with a reason to return to work: pending approvals, stalled diagrams, summaries. The texts are written by your model; enabling the STORM_RETENTION_ENABLED variable. Available in builds after October 7, 2026.

Getting Started

  • Cloud model (quick start) — connect OpenAI / Anthropic / ProxyAPI following the instructions in "Enabling and Connecting a Model". No own infrastructure is required.
  • Data within the perimeter — deploy a self-hosted LLM on your GPU and connect Storm to its endpoint.