Conversation Search¶
Agora can search your entire conversation history — either by keyword matching or semantic (meaning-based) retrieval using embedding models.
Search Methods¶
Keyword Search¶
Fast, exact text matching. The search looks for literal occurrences of your query in message content.
Best for: - Finding a specific phrase or term - Quick lookups when you remember exact wording - Zero setup — works immediately
Limitations: - Misses synonyms and related concepts - No understanding of meaning
Semantic Search (RAG)¶
Uses embedding models to find messages by meaning, not exact words. A query for "how to set up the database" can find messages about "Room configuration" even if the word "database" never appears.
Best for: - Finding conversations by topic or theme - Broad queries where you don't remember exact phrasing - Discovering related discussions across different conversations
Requirements: - An embedding model must be configured (see Embedding / RAG) - Messages must be cached (embeddings generated)
Setting Up¶
1. Add an Embedding Model¶
See Embedding / RAG for detailed setup. You can use: - Remote models (OpenAI, Mistral, Voyage, Ollama, etc.) - Local models (GGUF files, fully offline)
2. Choose Search Methods¶
In Settings → Conversation Search:
| Setting | Description |
|---|---|
| Model Search Method | How the model searches when it calls the search_conversations tool |
| Manual Search Method | How the search bar in the conversations drawer works |
Set each to Keyword or Semantic (RAG).
3. Configure Search Scope¶
| Setting | Range | Description |
|---|---|---|
| Context messages per search hit | 4–32 | How many surrounding messages to include with each match (steps: 4, 8, 12, 16, 20, 24, 28, 32) |
| Max search results | 5–30 | Maximum number of matches to return (steps: 5, 10, 15, 20, 25, 30) |
| Similarity Threshold | 0.0–1.0 | RAG only: minimum similarity score for a match. Higher = stricter. Default: 0.5 |
4. Cache Messages¶
If using RAG, tap Cache to generate embeddings for all existing messages. Enable Auto-cache to keep the index updated automatically.
Using Search¶
Manual Search (Search Bar)¶
- Open the conversations drawer (hamburger menu :material-menu: or swipe right)
- Tap the search bar at the top
- Type your query
- Results appear below — tap any result to open that conversation at the matching message
Model-Initiated Search¶
When Access Past Conversations is enabled (Settings → Memory), the model can search your history autonomously:
You: "What did we decide about the API design last week?"
Model: [Searches "API design decision"]
"Last Tuesday, we decided to use..."
The search appears as a tool card in the conversation.
Similarity Threshold¶
The Similarity Threshold slider (0.0 to 1.0) controls how closely a message must match to be included in RAG results:
- Low (0.3–0.5): More results, may include loosely related content
- Medium (0.5–0.7): Balanced — good default
- High (0.7–0.9): Fewer results, only very close matches
Start with the default and adjust based on your results. If you get too many irrelevant matches, raise the threshold. If you miss relevant conversations, lower it.
Search Results Display¶
In the conversations drawer, search results show:
- Conversation title (or "Untitled")
- Matching message — the user or model message that matched
- Role label — User or Model
- Context messages — surrounding messages for context
Tap a result to open the conversation scrolled to the matching message.