61cd010aff
* docs: rewrite user guide and core concepts around the current product
Rewrites docs/2-CORE-CONCEPTS and docs/3-USER-GUIDE against the code at
v1.15.0: current UI labels from the en-US locale, insights as a
first-class concept ("Insights only" context, Summary in podcasts),
context levels scoped to notebook Chat, Ask's three stages and notebook
scope, the real Add Source wizard, statuses and failure messages, real
transformation and podcast presets and fields, and Default Model
Assignments. Removes invented features (tags, export/share, batch
transformations, page-number citations, podcast bitrate/progress/share,
one-notebook-per-source, permissions) and generic filler.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: address review (model needs per feature, privacy scope, insight embeddings, upload limit, data storage, retry, UI details)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: address review (generated encryption key, when speech-to-text is needed)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: point provider setup to the canonical flow, fix Add (N) and Auto-assign scope
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
4.6 KiB
Search and Ask - Finding What You Need
Search and Ask live on the Ask and Search page (sidebar, under Process). Both work across your whole knowledge base, or only the notebooks you pick. There is no separate search box inside a notebook.
You can also start either from Quick actions (Ctrl+K / ⌘K): type a query and choose Search results for "..." or Ask about "...".
| Search | Ask (beta) | |
|---|---|---|
| Gives you | A list of matching sources, insights and notes | One written answer with citations |
| Uses an AI model | Vector search uses the embedding model; text search uses none | Embedding model plus a language model for each stage |
| Best for | Finding a passage, term or document | Questions whose answer is spread across sources |
How both work under the hood is explained in AI Context & RAG.
Search
- Open Ask and Search and choose the Search tab.
- Type your query and press Enter.
- Adjust if needed:
- Search Type: Text Search or Vector Search.
- Search In: Search Sources and/or Search Notes.
- Notebooks: pick notebooks to limit the search, or leave all unchecked to search everything.
Results show how many were found; click a result to open the source, note or insight. Where available, a result also lists its Matches (the matching passages).
Text Search vs. Vector Search
| Text Search | Vector Search | |
|---|---|---|
| Matches | Exact words (with English stemming) | Meaning, by similarity |
| Covers | Source titles and content, insights, note titles and content | Source content, insights, note content (not titles) |
| Needs | Nothing | An Embedding Model, and only finds embedded content |
| Use when | You know the term, name or phrase | You know the idea but not the wording |
If no Embedding Model is set, the page says Vector search requires an embedding model. Only text search is available.
Tips:
- Text search works best with distinctive words: names, acronyms, technical terms.
- Vector search works best with a descriptive phrase ("risks of relying on a single supplier") rather than one word.
- A source you can find with text search but not vector search probably isn't embedded. Open it and use Embed Content.
Ask
- Open Ask and Search and stay on the Ask (beta) tab.
- Type your question.
- Optionally limit it to some Notebooks (leave all unchecked for your whole knowledge base).
- Press Cmd/Ctrl+Enter or click Ask.
While it runs, you'll see the stages as they finish:
- Strategy: the model's reasoning and the Search Terms it chose (up to five searches).
- Individual Answers: one partial answer per search, based on the top matches.
- Final Answer: the combined answer, with links to the cited items.
Click Save to Notebooks to keep the answer as a note (titled with your question) in one or more notebooks.
Requirements
- An Embedding Model must be set. Without one, the Ask tab says You can't use this feature because you have no embedding model selected.
- A Chat Model must be set; it is used for all three stages by default.
- Ask only sees embedded content (sources added with embedding enabled, notes and insights).
Choosing models
The tab shows Using Default Models or Using Custom Models. Click Advanced to open Advanced Model Selection and choose a Strategy Model, Answer Model and Final Answer Model, then Save Changes. The strategy step needs a model that reliably returns structured output; if it fails with The strategy model returned no search terms for this question..., pick a different Strategy Model or rephrase.
When to use Ask
- Use Ask when you don't know which sources contain the answer.
- Ask is single-turn. To follow up, open a notebook and use Chat with the sources Ask cited.
- Answers come only from matched chunks, so they can miss context that is spread thinly across a document. For close reading, use Chat with Full content.
Troubleshooting
| Problem | Try |
|---|---|
| No results in vector search or Ask | Check that an Embedding Model is set and the sources are embedded; rebuild from Advanced → Rebuild Embeddings if you changed the embedding model |
| Text search misses a word | Try another form of the word or a synonym; stemming is English-only |
| Too many results | Limit to specific notebooks, or turn off Search Sources or Search Notes |
| Ask answer is empty or fails | Pick other models under Advanced; reasoning models with small output limits can return nothing |