Horizon — AI-augmented personal knowledge hub.

Solutions

1. 01.Agent router: selection with a confidence level, manual override possible
2. 02.RAG over embeddings to bring in the right context
3. 03.Streaming responses with source citations
4. 04.Editable artifacts in a canvas
5. 05.Pinned, centralized LLM models (a single value to change to migrate versions)
6. 06.API key strictly server-side, database access behind RLS
7. 07.Cost-first: web search off by default, cheapest model suited to each task, prompt cache preserved