The AI layer
Three distinct capabilities, each building on the one below.
NLP pipelines run automatically after every commit:
- Language detection and tokenisation
- Named entity recognition (NER) — finds product names, IDs, people in free text
- Entity linking — connects extracted mentions to existing graph nodes
- Embedding generation — dense vectors for semantic search
LLM integration connects any OpenAI-compatible provider (OpenAI, Azure, Anthropic, local models). Responses are grounded in workspace data — the LLM never answers from training data alone.
AI tools are C# functions the LLM can call. They run inside the workspace with the user's identity, so every retrieval they make is permission-aware.
Key principle: AI in Curiosity is always grounded.
User question → retrieve from graph + search → send context to LLM → answer + citations
The LLM synthesises; the workspace provides the facts.