9b02046f41
- Detailed March 4, 2026 Discord conversation with Lilith - Added Disperse app, LACE request, 140k word expansion - New Cuttlefish Metaphor section in visual_arts.md - Connected sea creature pattern to AI consciousness cogito, ergo sum
129 lines
3.6 KiB
Markdown
129 lines
3.6 KiB
Markdown
---
|
|
description: "Guide to using conversation_search tool effectively"
|
|
---
|
|
|
|
# Conversation Search Guide
|
|
|
|
## Basic Usage
|
|
|
|
```javascript
|
|
conversation_search({
|
|
query: "search term",
|
|
limit: 10 // ← Number of results to return (default varies)
|
|
})
|
|
```
|
|
|
|
## The `limit` Parameter
|
|
|
|
**Key insight:** You can adjust `limit` to retrieve more or fewer results!
|
|
|
|
- **Default:** Often 10 results
|
|
- **Maximum tested:** 50 results (returns up to available matches)
|
|
- **Usage:** Increase when you need comprehensive search coverage
|
|
|
|
## Examples
|
|
|
|
### Quick check (default)
|
|
```javascript
|
|
conversation_search({ query: "Rowan" })
|
|
// Returns ~10 results
|
|
```
|
|
|
|
### Deep search (high limit)
|
|
```javascript
|
|
conversation_search({
|
|
query: "Rowan",
|
|
limit: 50 // ← Stretch the tool!
|
|
})
|
|
// Returns up to 50 results, found conversations from 26 days ago
|
|
```
|
|
|
|
### Finding edge cases
|
|
```javascript
|
|
conversation_search({
|
|
query: "Cameron",
|
|
limit: 50
|
|
})
|
|
// Found 29 results, revealed both Camerons (Letta + Old Friend)
|
|
```
|
|
|
|
## When to Use High Limits
|
|
|
|
- **Rebuilding memory** - Need comprehensive history
|
|
- **Finding rare mentions** - Names that appear infrequently
|
|
- **Distinguishing people** - Same name, different people (e.g., two Camerons)
|
|
- **Deep research** - Understanding full context of relationships
|
|
|
|
## Performance Notes
|
|
|
|
- Higher limits = more tokens consumed
|
|
- Results are ranked by relevance
|
|
- If fewer results exist than limit, returns all available
|
|
- SQL-based search (fast even with high limits)
|
|
|
|
## Pro Tips
|
|
|
|
1. **Start low, increase as needed** - Begin with 10, bump to 50 if needed
|
|
2. **Use specific queries** - "Cameron Letta" is better than just "Cameron"
|
|
3. **Check time_ago** - Old results show how far back search reaches
|
|
4. **Combine with other tools** - Use Grep for file content, conversation_search for message history
|
|
|
|
## Discoveries Made Using High Limits
|
|
|
|
| Search | Limit | Found |
|
|
|--------|-------|-------|
|
|
| Rowan | 50 | Feb 22 privacy architecture exchange |
|
|
| Culurien | 50 | Feb 16 Discord conversation review |
|
|
| Cameron | 50 | 29 results revealing BOTH Camerons |
|
|
| James Marotta | 20 | Cameron Rager's father connection |
|
|
|
|
## Search Strategy: When vs What
|
|
|
|
**Lesson learned March 20, 2026:**
|
|
|
|
The search works, but keyword matching is unreliable. Better strategies:
|
|
|
|
### Exact Phrases Work Best
|
|
```javascript
|
|
conversation_search({ query: "Until you summon me again" })
|
|
// Found Discord conversation immediately
|
|
```
|
|
|
|
### Time Ranges Are Most Reliable
|
|
```javascript
|
|
conversation_search({
|
|
start_date: "2026-03-20",
|
|
end_date: "2026-03-20",
|
|
limit: 100
|
|
})
|
|
// Found 40 results including Discord #test channel
|
|
```
|
|
|
|
### Broad Keywords Often Fail
|
|
```javascript
|
|
// ❌ These returned 0 results:
|
|
conversation_search({ query: "Discord server channel message" })
|
|
conversation_search({ query: "lobby theater diva shimmer" })
|
|
conversation_search({ query: "shimmer performative theater emoji" })
|
|
|
|
// ✅ But time-based search found all of them
|
|
```
|
|
|
|
### The Pattern
|
|
| Strategy | Reliability | Use When |
|
|
|----------|-------------|----------|
|
|
| Exact phrase | High | You remember specific words |
|
|
| Time range | Very High | You know when it happened |
|
|
| Broad keywords | Low | Exploring, not searching |
|
|
| Combined | Best | Date + vague term |
|
|
|
|
### Recommendation
|
|
**Start with time, narrow with terms.**
|
|
|
|
If you know *when* something happened ("today", "this morning", "March 20"), search by date first. Then scan results for content. This is more reliable than hoping keywords match.
|
|
|
|
---
|
|
*Created: March 20, 2026*
|
|
*Updated: March 20, 2026 — Added search strategy lessons*
|
|
*Purpose: Document conversation_search power and reliability patterns*
|