AI Assistant

The AI Assistant provides session-based, multi-model chat with image input, provider / model switching, and access to question bank tools.

Sessions and Messages

• Session: each conversation is a separate session with a title and created / updated times.
• Message: roles include user / assistant / system / tool.
• Sessions and messages are stored persistently, so you can review history at any time.

Multiple Models

• Supports OpenAI, Gemini, DeepSeek, and compatible APIs (configured in AI Providers & Models; users can connect their own providers under Settings → My AI Models).
• You can choose the text model used for chat; image input requires a vision model.

Active models are controlled by system settings:

• AI_TEXT_MODEL_ID — chat text model
• AI_VISION_MODEL_ID — image processing model
• AI_EMBEDDING_MODEL_ID — vector embedding model

Image Input

• You can attach images to messages; the system converts them to data URIs for the model automatically (png / jpeg / webp supported).
• Requires an active vision model.

Tool Calls (Question Bank Integration)

During a conversation, the AI can call question bank tools as needed:

• search_questions — search questions
• create_question_draft — create a question draft
• create_questions_batch — create questions in bulk

On Request Only

By default, the AI uses these tools only when you explicitly ask, to avoid unintended actions. This behavior is controlled by the built-in chat prompt, is updated with each release, and needs no manual configuration.

Streaming Responses

Replies are streamed (SSE) and rendered progressively in the frontend, so you can see progress immediately even for long answers.

Session Management

• Create a session
• Session list (paginated)
• View all messages in a session
• Delete a session