Skip to main content

AI assistant

PlanDesigner plandesigner

The Studio assistant is a chat panel on the Data Designer, the Paths editor and the workspace. It, and every AI button on the design surfaces, talks to an Ollama server your installation runs — not to the Suite's LLM gateway, which serves Scribe. When Ollama cannot be reached, the assistant falls back to an offline responder and says so.

Open it​

The floating AI Chatbot button, or Ctrl / ⌘+Shift+A. The panel slides out or goes full screen. In the workspace it is scoped to the working set and names its scope — "Answering about customers/ — 2 schemas, 4 operations on the canvas"; on the classic canvas its context is the version's classes and paths.

Models​

The model picker lists the generative models Ollama has (embedding models are filtered out), with a Tenant default option — "Remember this model as the default for every project in this tenant".

  • "Could not reach Ollama. Using the offline assistant." with Retry.
  • "No generative models in Ollama. Pull one (for example Qwen 2.5 or Llama 3.2) or continue with the offline assistant."

The conversation​

The empty state — How can the Studio assistant help? — suggests prompts such as "Create a User class with email, password hash, created date, and roles array" and "Best practices for this workspace". The composer reads Ask the Studio AI…, with Stop generating while it streams. The toolbar has New, Export (Markdown) and Clear; Past conversations are searchable, deletable and kept on your device. The What the assistant can see chip lists the project, domain, version, selection, classes, paths and properties in its context.

Act on a reply​

ActionDoes
Create this classOpens Preview class schema — Refine with the assistant or Continue to create
Add these properties, Apply to current class, Open AI property suggestionsWrites suggested properties into the selected class
Preview import → Apply importFor an OpenAPI document in the reply
Import this diagramFor a Mermaid class diagram
Copy JSON/YAML, Regenerate response, helpful / not helpfulOn every reply

AI buttons​

The same server backs the buttons on the surfaces: Suggest properties with AI and Suggest types with AI in the property dialog ("Requires Ollama when connected."), Suggest description and Suggest properties in the workspace inspector, Create Class with AI in the class dialog, and the operation docs button on the Paths editor. Behind them are server tasks for class skeletons, property suggestions and types, property and class descriptions, examples, operation descriptions, schema improvements and canvas layout recommendations.

For operators​

Point the studio at Ollama with OLLAMA_BASE_URL (default http://localhost:11434); the optional OLLAMA_CHAT_CACHE_* variables size the response cache. See the environment reference.