Build agents
Data preparation
Start from business questions and use a desktop AI assistant to clarify questions, design tables, and check data gaps.
When you are not sure which data to prepare, start from business questions: first clarify the questions to answer and the metric definitions, then design the tables you need, and finally check what the data already in AskTable is missing.
These three steps run in your own desktop AI assistant (such as WorkBuddy, Qwen Work, or Codex), guided by three skills. They are skills for the desktop assistant, different from analysis skills in agents. The skills produce a preparation plan only and do not change data sources.
Entry: “More” → “Data preparation”.
Three steps
On the “Data preparation” page, click “Copy instructions” in order and paste them into the same conversation in your desktop assistant. The instructions include the skill package download address; the assistant installs the skill before running it.
| Step | Skill | Output |
|---|---|---|
| 1. Clarify the questions | asktable-discover | Question list, metric definitions, items to confirm |
| 2. Design the tables you need | asktable-model | Table structure, data dictionary, mapping between questions and data |
| 3. Check AskTable data gaps | asktable-check | Answerable questions, data gaps, and a plan to fill and recheck them |
Step 3 reads the data sources you can access in AskTable through the AskTable Skill, so install and sign in to the AskTable Skill in the assistant first. See CLI.
What next
- Fill data gaps in AskTable according to the list: Connect data sources, Table connectors, or AI collection tables.
- Run step 3 again to recheck.
- When the data is complete, build an agent and start analyzing.
Make good use of it
- Start step 1 from the decision to make, such as “decide which stores adjust staffing next quarter”, not from existing tables.
- Complete all three steps in the same conversation; each step uses the previous step's results.
- Once definitions are confirmed in step 1, write them into the agent's business documents for later questions.