Skip to main content
Quark retrieves the closest matching chunks from Qdrant, hydrates the prompt with short-term session memory and long-term user memory, and sends the grounded prompt to the LLM. Every answer is constrained to your ingested content. When no chunk scores above the similarity threshold, Quark responds with:
I could not find any relevant notes for your question.

Querying from the CLI

Type any text that does not start with : and press Enter.
What are the main findings in section 3?
1

Submit your question

Type the question at the prompt. The active session label is visible in the prompt box.
2

Wait for the answer

A spinner labelled Thinking is shown while Quark searches the vector store and calls the LLM.
3

Read the answer

The answer is printed as rendered Markdown, followed by source citations. The exchange is saved to the session’s local chat log automatically.
Use :ingest with a specific institution tag before querying to scope retrieval to that document set. The institution filter persists for the duration of your session.

Build docs developers (and LLMs) love