# Conversations Query data in natural language, inspect the analysis process, and ask follow-up questions. 使用指南 · 模块: product · 使用者 · 部署形态：资料未说明 · 已确认产品版本：资料未说明 · 来源：现有文档 Conversations are AskTable's main entry point: click “New conversation” in the main navigation on the left, select your data source, and ask a question in one sentence to get a result with its sources. You do not need to write SQL or build a report first. Use conversations for ad hoc queries, exploring business questions, and follow-up questions. ## Before you start | Item | Requirement | | ------------ | ----------------------------------------------------------- | | Permissions | Your role can access the data sources being used | | Prerequisite | A data source is connected and metadata parsing is complete | | Entry point | Main navigation on the left → New conversation | ## Decide three things before asking The three selectors below the input box and the shortcuts beside it define the scope of your question: | Setting | Location | Description | | ----------------- | --------------------------------------------------------------- | ------------------------------------------------------------------------------------------------- | | Data source | Data source selector below the input, defaulting to all sources | Keep all sources selected to ask across sources | | Agent | Agent selector below the input | Once selected, the data source selector becomes read-only; the agent decides which sources to use | | Role | Role selector below the input | Determines row-level data scope and variable values; select “No role” when unnecessary | | Skill | “+” menu to the left of the input → “Skills” | Attach an analysis method, such as attribution or drill-down | | Attachment upload | “+” menu to the left of the input → “Upload Excel/CSV” | Supports Excel and CSV; does not depend on the Python interpreter | | Build a dashboard | “+” menu to the left of the input → “Build Dashboard” | Enter dashboard-building mode directly | | Run mode | “+” menu to the left of the input → “Fast mode” | **Off by default, meaning deep thinking**; turn it on for faster but less detailed analysis | Changing these settings after a conversation has started opens “Modify conversation configuration”, explaining that the change will create a new conversation. ## The question workflow **Send a question** Write your question in the input box. You can also click “Upload attachment” or “Upload Excel/CSV” in the “+” menu to attach an Excel or CSV file directly. By default, Excel files can be up to 10 MB, with at most 20 sheets per file and at most 200 fields (columns) per sheet; CSV files can be up to 20 MB with at most 50 fields (columns). Uploading does not depend on the Python interpreter. **See which tools are used** The message moves through thinking and analyzing states. Actual tool calls appear during the process, labeled in Chinese with these meanings: Query data, view data tables, create data artifacts, execute Python, view result rows, create charts, read skills, ask the user, and search the web. The end of each answer shows the number of tool calls and messages used. **Answer clarification questions when needed** If the agent determines that conditions are missing, it asks a question and lists options. Select your answers one by one, then click “Submit”. Use “Previous” to revise an answer, “Skip” to let the agent decide, or “Other” to enter your own answer. The user clarification switch controls this behavior; disabling it removes these follow-up questions from the agent. **Review the results** The results area has “Table” and “Chart” tabs. Chart types include line, bar, and pie charts. The table shows its current row and column counts and supports filtering, sorting, and row search. Use “Edit chart” to adjust chart settings. **Take the results with you** | What you need | What to use | | ---------------------------------------- | ---------------------------------------------------- | | A file | “Download CSV” | | A reusable data table | “Save to Tables”; find it later under “View Table” | | Content to paste into Excel or Word | “Copy table” | | Content to paste into PowerPoint or Word | “Copy as image” or “Copy as SVG” | | A result to show someone else | “Share conversation” → enable the public link switch | Public links are read-only. Anyone with the link can view the conversation. ## Follow-up questions and long conversations - **Just keep asking.** Add filters, dimensions, a different period, or a different sort order in the same conversation; the agent uses the existing context. - **Quote a passage.** Select content and use “Quote” to include it in the next question, avoiding ambiguous references. - **Jump between turns.** Use conversation turn navigation to jump to a particular turn, or “Jump to latest” to return to the end. - **Monitor context usage.** The top of the page shows the percentage of context used. When it is nearly full, run the compression command to free model memory. After success, the conversation is marked as compressed. - **Find past conversations.** “All conversations” in the sidebar supports name search and type filtering. Types include chat, dashboard, embed, scheduled task, and simple query. ## Key constraints | Item | Value | Where to change it | | ------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------ | | Default run mode | Deep thinking (“Fast mode” is off by default) | “+” menu → “Fast mode” | | Per-turn processing limit | Processing pauses when the limit is reached; click “Continue” to resume | — | | Single-message length | Limited; messages exceeding the limit are rejected | — | | Charts | Depend on the chart visualization switch | Agent configuration → “Capabilities” | | Attachment uploads | Excel and CSV; defaults are Excel up to 10 MB per file, at most 20 sheets per file, and at most 200 fields (columns) per sheet; CSV up to 20 MB per file and at most 50 fields (columns); does not depend on the Python interpreter | — | | Web search | Depends on the web search switch and prior configuration by an administrator in system settings | Agent configuration → “Capabilities” | ## Troubleshooting | Symptom | How to identify it | What to do | | ---------------------------------------- | --------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- | | Answers are less detailed than usual | “Fast mode” is on | Turn it off to return to default deep thinking | | No response for a long time | No new tool calls for more than 1 minute | Wait; click “Continue” if processing pauses | | Cannot send a message | Sending is rejected with a project-locked message | Contact an administrator about licensing | | Cannot find a previous conversation | “All conversations” is empty or search finds nothing | Check whether the type filter or last-active time range is too narrow | | Numbers do not match the business system | Expand the tool call to inspect the SQL actually executed | Check the time range and granularity; improve [table and field descriptions](https://docs.asktable.com/en/data/semantics) for definition issues | | No chart appears | Chart visualization is disabled for the agent | Enable it in agent configuration | ## Next steps - Attribution, drill-down, or multidimensional comparison: use an [analysis canvas](https://docs.asktable.com/en/canvas) - Ongoing monitoring of results: add them to a [data dashboard](https://docs.asktable.com/en/dashboard) - Let your team query the same knowledge layer from Feishu: see [Feishu bot](https://docs.asktable.com/en/integrations/feishu) - Use an API or MCP: see [Integration options](https://docs.asktable.com/en/integrations/overview)