QueryTune
Best Practices for Context & Tuning
To get the most out of QueryTune, providing the right context to the AI is just as important as choosing the right model.
🧠 Recommended Models
| Model Name | Type | Best For | Suggested Temp |
|---|---|---|---|
qwen2.5-coder:7b |
Local | Default choice. Balanced speed and excellent SQL logic. | 0.1 |
deepseek-r1:8b |
Local | Reasoning-based analysis. Great for deep explanations. | 0.6 |
llama3.1:8b |
Local | SQL generation and reliability. | 0.2 |
qwen3-coder-next |
Cloud | Top cloud coder, Ollama interface. SQL opt + agentic | 0.2 |
gpt-4o |
Cloud | Very fast and reliable generalist for complex tasks. | 0.0 |
claude-3.7-sonnet |
Cloud | Exceptional at following complex refactoring rules. | 0.0 |
Pro Tip: For SQL optimization, always keep the temperature low (0.0 - 0.2) to ensure syntactic correctness, except for “Reasoning” models like DeepSeek-R1 which perform better with a slightly higher temperature (0.5 - 0.7).
🎯 How to Provide Context
AI models are not psychics; they need to know your database structure to suggest valid optimizations.
1. Surgical Context
Provide only the DDL (CREATE TABLE) of the tables actually involved in the query. Adding unrelated schemas increases “noise” and the risk of hallucinations.
2. Data Statistics
Explicitly state table sizes in the Context area. Example: `“Table users has 10M rows, Table orders has 500 rows.”* This is critical for the AI to suggest the correct JOIN order and execution strategy.
3. Identify Constraints
Inform the AI about specific limits:
- “Do not use window functions (not supported in my DB version).”
- “I cannot add new indexes, only refactor the SQL.”
4. Context Window
For long or complex queries, prefer models with at least 32k context (like Qwen 2.5 or Claude) to avoid the AI “forgetting” the beginning of the prompt.
🔐 Privacy Note
Only local models (via Ollama) guarantee complete privacy. When using Cloud models (OpenAI, Anthropic), your query and context are sent to their servers. Never include sensitive data or real production secrets in your context.