Overview
Whim’s search lets you find information across your entire workspace — task names, AI conversations, and shell command history. It combines traditional keyword search with semantic understanding to surface relevant results even when exact terms don’t match.Search modes
Whim supports three search modes, selectable when performing a search:| Mode | How it works | Best for |
|---|---|---|
| Keyword | PostgreSQL full-text search with ranking | Finding exact terms, commands, error messages |
| Semantic | Vector similarity using embeddings | Finding conceptually related content, natural language queries |
| Hybrid | Combines keyword + semantic with Reciprocal Rank Fusion | General-purpose search (default) |
Hybrid mode is the default and works best for most searches. It combines the precision of keyword matching with the flexibility of semantic understanding.
What you can search
Task names and prompts
The sidebar search finds tasks by matching against their name and prompt content. This is the fastest way to locate a specific task when you remember part of its name or instructions.Conversations
Search through the full AI conversation history across all tasks. This includes everything the agent said, every file it discussed, and every decision it explained. Useful for finding past solutions or understanding how a problem was approached.Shell history
Search through terminal command history — every shell command run inside task containers. This is invaluable for finding that specific command sequence you ran last week, or discovering which task ran a particular deployment script.Search filters
Refine your results with these filters:| Filter | Description |
|---|---|
| History type | Search across all content, conversations only, or shell history only |
| Task scope | Limit search to specific tasks |
| Date range | Filter by start and/or end date |
| Match threshold | Adjust the minimum similarity score for semantic results (0–1) |
Using search
Sidebar search
Type in the search box at the top of the sidebar to quickly filter the task list. This searches task names, prompts, and conversation content, surfacing tasks whose content matches your query. Results are ranked by relevance:- Name/prompt matches — highest priority, shown first
- Conversation matches — tasks where the AI discussed related topics
- Shell matches — tasks where matching commands were run
Conversation search panel
For deeper exploration, open the conversation search panel within a task view. This provides:- Full search across conversation and shell history
- Preview snippets showing the matching context
- Links directly to the relevant task and conversation point
- Metadata including task status, tags, branch, and who created the task
Result ranking
Search results are ranked by a combination of:- Keyword relevance — how closely the text matches your query terms
- Semantic similarity — how conceptually close the content is to your query (when using semantic or hybrid mode)
- Recency — more recent results may be prioritized
Semantic search may fall back to keyword-only mode if the embedding service is temporarily unavailable. When this happens, you’ll see an indicator that semantic results are limited.
Tips for effective search
Use keywords for precision
Search for exact error messages, function names, or file paths when you know the specific term.
Use natural language for discovery
Ask questions like “how does authentication work” to find conceptually related conversations.
Filter by history type
Switch to “shell” when looking for a command, or “conversations” when looking for a discussion.
Narrow by date
If you remember roughly when something happened, use date filters to reduce noise.