AI-powered search
Ultrathink's AI-powered search lets you query your knowledge base using natural language. Instead of relying on exact keyword matches, you can ask questions the way you would ask a colleague, and the AI will find relevant entries and generate an answer.
How AI search works
When you submit a search query, the AI:
- Interprets your intent: Understands what you are looking for, even if you do not use the exact words stored in your entries
- Searches semantically: Finds entries that are conceptually related to your query, not just those containing the same keywords
- Generates an answer: Produces a written response that synthesises information from your matching entries
- Cites sources: Links back to the specific entries that informed the answer
This means a search for "that article about making APIs faster" can find an entry titled "REST API Performance Optimisation Techniques" even though the words do not match exactly.
Accessing AI search
There are two ways to search in Ultrathink: quick search and AI search.
Quick search
Press Ctrl+K (or Cmd+K on Mac) from anywhere in the app to open quick search.
- Searches titles and content using text matching
- Results appear instantly as you type
- Use arrow keys to navigate and Enter to open an entry
- Best for finding a specific entry when you know part of its title or content
AI search
Navigate to Search in the sidebar for the full AI search experience.
- Accepts natural language questions
- Returns an AI-generated answer plus matching entries
- Supports entity filtering
- Best for exploratory queries and finding information across multiple entries
Quick search vs AI search
| Feature | Quick search | AI search |
|---|---|---|
| Access | Ctrl+K / Cmd+K from anywhere | Search page in the sidebar |
| Query type | Keywords and phrases | Natural language questions |
| Matching | Text-based (titles and content) | Semantic (meaning and context) |
| Results | List of matching entries | AI-generated answer with source entries |
| Speed | Instant | A few seconds (AI processing) |
| Filtering | None | Entity type filtering |
| Best for | Finding a known entry quickly | Exploring your knowledge base |
Use quick search when you know what you are looking for. Use AI search when you want to discover connections or get an answer that spans multiple entries.
Natural language queries
AI search understands questions and requests written in plain language. You do not need to use special syntax or boolean operators.
Example queries
| Query | What it finds |
|---|---|
| "What did I save about React performance?" | Articles, notes, and conversations related to React performance optimisation |
| "Meeting notes from last week" | Entries with meeting-related content from recent dates |
| "John's recommendations for the API" | Entries mentioning John that contain API-related suggestions |
| "That article about serverless" | Serverless-related links and snippets |
| "Project deadlines coming up" | Tasks and projects with approaching due dates |
| "Comparisons between AWS and Azure" | Entries discussing cloud platform trade-offs |
| "What has Sarah said about the redesign?" | Conversation transcripts and notes involving Sarah and the redesign project |
Tips for writing queries
- Be conversational: Write as if you are asking a person, not a search engine
- Include context: Mention people, projects, or timeframes when relevant
- Ask specific questions: "What are the pros and cons of GraphQL based on my saved articles?" works better than just "GraphQL"
- Try different angles: If one query does not find what you need, rephrase it from a different perspective
AI-generated answers
When you run an AI search, the response includes two parts:
The answer
A written summary that synthesises information from your matching entries. This is not a simple copy-paste of your content; the AI reads across multiple entries and constructs a coherent answer to your question.
The answer may include:
- Key points from relevant entries
- Comparisons or contrasts between sources
- Dates, names, and specific details extracted from your knowledge
- References to which entries contributed to the answer
Source references
Below the answer, you will see the entries that the AI drew from. Each source shows:
- The entry title (clickable to open in the detail panel)
- A brief excerpt showing why it matched
- The entry type and date
Source references let you verify the AI's answer and dig deeper into any specific entry.
Entity filtering
Narrow your search results by entity type to focus on a specific kind of content.
Available filters
| Filter | What it includes |
|---|---|
| All | Every entry type (default) |
| Projects | Entries classified as projects |
| Tasks | Entries classified as tasks |
| Knowledge | Entries classified as general knowledge |
When to use entity filters
- Projects filter: When you want to find project-related information without noise from general knowledge entries
- Tasks filter: When looking for action items, to-dos, or work with deadlines
- Knowledge filter: When searching for reference material, articles, or research
Select a filter before or after running your query. The AI regenerates the answer based on the filtered set of entries.
Search result quality
The quality of AI search depends on how well your entries have been processed. Here are factors that affect results:
| Factor | Impact on search |
|---|---|
| AI processing complete | Entries that have finished all four pipeline steps are fully searchable |
| Rich notes | Entries with detailed notes provide more context for matching |
| Accurate classifications | Correct entity types and topics improve filtered searches |
| Descriptive titles | Clear titles help the AI understand what entries are about |
| Populated topics | Entries tagged with topics are easier to find when searching by subject |
Common search patterns
Finding saved content
"Where did I save that article about design systems?" "What bookmarks do I have about TypeScript?"
Synthesising knowledge
"What do I know about our competitors' pricing?" "Summarise my notes on the product launch."
People-related searches
"What meetings have I had with the marketing team?" "What has David shared about the infrastructure migration?"
Time-based queries
"What did I capture last week?" "Recent entries about the Q2 roadmap."
Decision support
"What are the arguments for and against switching to Postgres?" "Based on my research, which framework should I choose?"
Troubleshooting
| Problem | Solution |
|---|---|
| No results for a query you expect to match | Check that the relevant entries have completed AI processing (view AI Progress in the detail panel) |
| Answer does not include a specific entry | The entry may not be semantically related enough; try mentioning specific keywords from that entry |
| Results are too broad | Use entity filtering or add more specific terms to your query |
| Results are too narrow | Broaden your query or remove specific constraints |
| Answer seems inaccurate | Check the source references to verify; the AI synthesises from your content, so the source entries determine accuracy |
| Search is slow | Complex queries with large knowledge bases may take a few extra seconds; this is normal |