The five query modes
The chip in the search bar sets how a question is taken. The modes come in two groups.
| Mode | Shortcut | Needs | What it does |
|---|---|---|---|
| AI search | |||
| Auto | ⌘1 | Claude Code or Codex signed in; (the embedding model, for text routes) | The AI routes each question: to Cypher, to vector search, to vector search then the matches’ connections, or to Cypher then vector search within the result. With nothing embedded there’s nothing to route and Auto is plain Cypher. |
| Text to Cypher | ⌘2 | Claude Code or Codex | The AI turns your question into one Cypher query and runs it. |
| Vector | ⌘3 | the embedding model (310 MB, once); no account | Your words are embedded on this Mac and the closest passages of embedded text are found. ::Type:: picks where. |
| Raw query | |||
| Raw Cypher | ⌘4 | nothing | Runs the Cypher you type, read-only, exactly as written, with completion as you type. |
| Raw Text | ⌘5 | nothing | Finds the nodes and relationships whose name or text contains your words; commas for several; ::Type:: picks where. |
The chosen mode is remembered between launches. Each is described on its own page: Raw Text, Raw Cypher, Text to Cypher, Vector search, Auto.
Which to use
- You know the node’s name, or a word in it: Raw Text. It’s instant, exact, and needs nothing.
- You remember the idea but not the words: Vector. It finds passages that mean the same thing.
- The question is about structure (counts, who connects to what, before 1950, the most cited): Text to Cypher, or Auto.
- You’d rather let the AI decide, and some text is embedded: Auto.
- You write Cypher: Raw Cypher. The completion knows your shape.
What each sends off the Mac
| Mode | Leaves the Mac |
|---|---|
| Raw Text, Raw Cypher, Vector | nothing |
| Text to Cypher | the question and the shape (type and property names, which property is each title) |
| Auto | the same, to the routing model; then whatever the chosen route needs |
What the trail shows
Every mode writes a trail of steps under the bar: what ran, with whom, how long it took, and what it found. The steps per mode:
- Raw Text: Searching names and text · on this Mac → Searched names and text · 12 matches · 3 ms.
- Raw Cypher: Your query · no AI step (the query, expandable) → Running on your graph → Ran on your graph · 12 rows · 4 ms.
- Text to Cypher: Translating to Cypher · Sonnet (with the model’s thinking as it streams) → Translated to Cypher · Sonnet · 3.2 s (the query) → Ran on your graph; or Used cached query · Sonnet · no API call; or Repairing the query; or The database rejected the query; or No query for this.
- Vector: Running vector search · on this Mac → Ran vector search · 8 matches · 2 ms.
- Auto: Routing the question · Sonnet → Routed to Cypher / Routed to vector search / Routed to vector search, then connections / Routed to Cypher, then vector search → the steps of that route.