Summary
The Model tab controls which AI models answer chats, database type, optional BYOK OpenAI key, retrieval limits, and entity-aware RAG routing.
Prerequisites
- Edit chatbot → Model tab
Settings reference
| Setting | Default | Range / options | |
|---|---|---|---|
| LLM Model | Platform default | List from server | |
| Data Base Type | Standard Database | Standard \ | Enhanced Database |
| Simple LLM Model | Same family | Used when simple mode on | |
| Use simple model first | Off | Checkbox | |
| Use your own OpenAI API key | Off | Checkbox | |
| OpenAI API Key | Empty | sk-... when BYOK on | |
| Vector Search Limit | 10 | 3–20 | |
| Message History Limit | 20 | 2–200 | |
| Entity-aware RAG routing | Varies | On = hybrid entity + semantic search |
Data Base Type
| Type | Description |
|---|---|
| Standard Database | Vector search (default; used by WordPress Connector) |
| Enhanced Database | LightRAG-enhanced knowledge graph path |
Changing type may require rebuild and longer processing.
Simple Model Mode
When Use simple model first is enabled:
- Lightweight model tries to answer first (lower cost)
- Falls back to advanced LLM Model if needed
When disabled, only the advanced LLM Model is used.
API Key Mode (BYOK)
When Use your own OpenAI API key is enabled:
- Your key is used for this bot’s OpenAI calls
- You are billed by OpenAI directly
- Saved key is retained if you temporarily disable BYOK
When disabled, the platform default key is used (included in beta).
Vector Search Limit
Number of knowledge chunks retrieved per user question (3–20). Higher = more context but slower and costlier.
Message History Limit
How many recent user/assistant turns the AI sees (2–200). Full history is still stored in Conversation History.
Entity-aware RAG routing
When enabled, the bot detects entities in questions and uses hybrid search (semantic + entity chunks). When disabled, legacy semantic-only search is used.
What triggers a rebuild
| Action | Rebuild needed? |
|---|---|
| Change LLM / simple model settings | No |
| Change database type | Often yes — wait for processing |
| Change vector/history limits | No |
| Toggle entity-aware RAG | May require rebuild for index |
Limitations
- Enhanced database not available on all plans (beta: check status messages)
- Invalid OpenAI key causes chat errors immediately
- Model list depends on server configuration
Troubleshooting
Symptom: “Invalid API key” errors
Cause: BYOK key wrong or expired.
Fix: Re-enter key; disable BYOK to use platform key.
Symptom: Slow responses
Cause: High vector limit or enhanced DB.
Fix: Lower Vector Search Limit; enable simple model mode.