What is AI Observation?
Make the "thinking process" of AI visible, accurate, and easy to modify.
What is AskTable AI Observation Service?
While using AskTable, have you encountered these problems?
- The query results are inaccurate but you don't know where the problem lies?
- Some user queries take a long time or even result in errors, and you don’t know how to troubleshoot?
- You want to optimize a certain type of question uniformly but can’t understand how AI “thinks”?
The AI observation service was born for this purpose. It can fully record and visualize the "reasoning chain" and "thinking path" of AskTable when processing each query — including model call status, key steps, generated content, time consumption and resource usage, error points, etc., turning AI from a "black box" into a "transparent glass box."
Why do you need it?
| Feature Module | How it helps you |
|---|---|
| 🌐 Full-chain Visualization | Gradually unfold each layer of reasoning process from user questions to final results, helping you understand how the system thinks. |
| 🔍 Precise Problem Diagnosis | Supports step-by-step troubleshooting of bottlenecks, high latency, errors, etc., suitable for operations personnel and data engineers to troubleshoot independently. |
| 📈 Query Effect Optimization | Understand the logic of AI in selecting tables, generating SQL, filtering data, etc., and find optimization points. |
| 🧠 Transparent and Trustworthy AI | Show the "explainability" of AI results to business stakeholders and management to build trust. |
Example Screenshots: Transparent Intelligent Agent "Thinking Chain"
The following figure shows an AI observation log snapshot over a period of time:
The following two figures show the real reasoning process of a user query from two perspectives in the backend:
- The time consumption, input, and output of each model call are clearly visible.
- Which steps depend on external data? Did they hit the cache? Did any errors occur?
- The generated SQL and summary results can be directly compared with the database execution results.
Differences Between Versions
| Deployment Version | Does it include observation service | Notes |
|---|---|---|
| Single-machine Experience Edition | ❌ Not supported | To reduce resource usage, this version only retains basic functions. |
| Local Deployment Enterprise Edition A | ✅ Supported | ✅ Recommended version, complete features and high cost-effectiveness. |
| Local Deployment Enterprise Edition B | ✅ Supported | Supports offline operation and deep customization, suitable for large customers. |
Why is Local Deployment Enterprise Edition A your top choice?
- ✅ Has complete observation capabilities to ensure AI controllability, adjustability, and explainability.
- ✅ Provides private deployment + data desensitization and encryption to ensure data privacy and compliance.
- ✅ Delivered and launched within 3 days + exclusive technical support team support.
- ✅ High cost-effectiveness: Starting at $9,000/year, including a complimentary computing power package, suitable for SME budgets.
Recommended Scenarios
- ✅ Data-intensive departments (such as data platforms, IT departments, etc.)
- ✅ Technical teams that need to review or optimize AI outputs.
- ✅ Teams with internal SQL or BI capabilities who want to further improve efficiency and transparency.