Search & Chat¶
Hybrid Search¶
Knowmarks uses hybrid search that combines two ranking signals:
- Semantic search — Vector embeddings capture meaning. Searching "ML deployment" finds articles about "machine learning in production" even if they don't contain the exact words.
- Full-text search — SQLite FTS5 handles exact keyword matching, including prefix matching and phrase queries.
Results are merged using Reciprocal Rank Fusion (RRF), which produces a single ranked list from both signals.
CLI¶
km find "distributed systems consensus"
Dashboard¶
Type in the input bar at the bottom of the screen. Results appear with relevance explanations showing why each item matched (e.g., "keyword: consensus", "semantic: distributed systems patterns").
Conversational Search¶
When you phrase your query as a question, Knowmarks uses an LLM to generate an answer grounded in your saved content:
What design patterns work best for event-driven architectures?
The answer cites specific items from your collection, so you can trace every claim back to a source.
Note
Conversational search requires an LLM provider. Without one, all queries use standard hybrid search.
Multi-turn Conversations¶
Conversations persist in the sidebar. Click a previous conversation to continue it with follow-up questions. The LLM maintains context across turns.
Chat Actions¶
The chat interface can also execute actions on your collection:
- "Create a collection about Rust async" — Creates a curated collection with relevant keywords
- "Add this to my DevOps project" — Associates items with a project
- "Delete all items from example.com" — Bulk operations with a preview step
Actions show a preview of what will happen before executing. You can refine, confirm, or cancel.
Scoped Search¶
By cluster¶
Browse a cluster from the sidebar, then search within it. Results are filtered to items in that cluster.
By project¶
Open a project view, then search. Results are scoped to the project's associated items.
By time¶
From the CLI:
km find "react hooks" --since 30d
Query Expansion¶
When an LLM provider is configured, Knowmarks can expand your query before searching. A query like "k8s networking" gets expanded to include related terms like "Kubernetes", "CNI", "service mesh", and "pod networking" — improving recall without losing precision.
Query expansion happens automatically for conversational queries. For standard searches, it activates when the initial result set is small.