Common Features

Questions, conversation contents, attachments, and query results may be sent to the model service configured by the administrator. Use only data your organization permits sending to that service. Do not provide passwords or API keys.

1. Query and Analyze Data

Describe the task directly, for example:

Find the five devices with the highest average temperature yesterday. State the time range and temperature unit.

Check the time range, objects, and units in the response. A "Results truncated" message means only part of the results is displayed. For complete statistics, ask the assistant to aggregate before answering:

Calculate statistics over all matching data. Do not draw conclusions from only the few rows currently displayed.

2. Generate Charts

After receiving query results, ask:

Plot these devices' average temperatures as a bar chart.

Or:

Query this device's hourly average temperature yesterday and plot a line chart.

A chart card appears in the response. Check its time range, axes, and units.

3. Generate Reports

Complete the required queries, then ask:

Generate a daily operations report using the data already queried in this conversation. Include the coverage, main metrics, anomalies, and recommendations. Mark items without supporting data as requiring confirmation.

Get the report: Click Copy Markdown or Download .md on the deliverable card. Markdown is a text format and can also be opened in a text editor.

Check before use: Confirm that the report's conclusions agree with the actual query results.

4. Answer Questions Using Knowledge Base Procedures

The knowledge base contains operating procedures, documentation, and other materials uploaded by administrators. For example:

Explain the inspection steps after a temperature limit is exceeded, using the equipment operating procedures. Cite the document name and section.

Check sources: Answers should cite document names, pages, or sections. For more context, ask the assistant to read the surrounding material before explaining. If the administrator configures automatic retrieval, the engine searches the knowledge base once at the start of each turn, with matched sources shown under Knowledge Base Search in the process details. Administrators can disable automatic retrieval; in that case, explicitly mention the knowledge base when needed.

If no material is found:

1. Try complete keywords used in the document. For English, prefer a full word such as temperature; temp may not match. Names with underscores are split for retrieval, so iot can also match demo_iot.
2. If the material is still not found, ask an administrator to check document status and the assistant's knowledge base settings.

5. Export Query Results

1. After a query, ask "Export these results as CSV." Export is read-only and requires no approval.
2. Check the filename, row count, size, and columns on the download card, then download the file.

Limits: By default, each file contains at most 100,000 rows and 50 MB. Content beyond the limit is omitted and the file is marked as truncated. Text starting with =, +, -, or @, except pure numbers, is prefixed with a single quote to prevent spreadsheet software from interpreting it as a formula.

Access and cleanup: Only the user who owns the conversation can download the file. It is removed when the conversation is permanently deleted.

6. Check Database Status

Ask "How is the database running?" or "Which node has a problem?" The assistant checks connectivity, version, activation, nodes, regions, synchronization tasks, running queries, system variables, and disks, and reports each result. These checks are read-only and require no approval.

7. Forecasting

Before starting:

• A Timecho AI key must be configured, and the forecasting tool must be enabled for the current agent. Contact a platform administrator if needed.
• Use one time series. Filter by device first if the data contains multiple devices.
• Sort data by time in ascending order, with approximately equal intervals and at least 16 rows.
• By default, at most the last 2,000 rows are sent, and the forecast horizon is 1–720 steps.

Run a forecast:

1. Ask, for example, "Forecast this device's temperature for the next hour."
2. Check the columns, row count, and destination service on the approval card, then approve.
3. Review the forecast table and the history/forecast curve.

Data transfer and approval: Forecasting sends the time and target columns from query results to the configured Timecho AI service. An approval card therefore appears even for data obtained by a read-only query. Each transfer is recorded as Data Egress in Audit Center. Select Automatically approve similar operations in this conversation to avoid approving every subsequent forecast in that conversation.

8. Use Documents and Scanned Files

Upload and ask:

1. Click Document below the input box, or drag a file into the box.
2. Wait for upload and parsing to finish. Check the filename, character count, and page count on the attachment bar.
3. Enter your request, such as "Summarize the routine inspection steps in this document and cite page numbers," then send it.
4. Check the document names and page numbers in the answer. For a long document, ask the assistant to continue reading subsequent sections.

Remove an unsent attachment using ×. Switching conversations discards unsent attachments.

Formats and limits:

Item	Default requirement
Formats	PDF, Word (.docx), TXT, Markdown, CSV; save legacy Word .doc files as .docx first
File size	At most 20 MB per file
Count	At most five documents per message and 20 retained documents per conversation
Document text	At most 500,000 characters per document; excess content is truncated and marked
Total upload size	Also subject to the total request size limit; retry in batches or split the files if necessary

CSV uploads: Uploading CSV in a conversation allows the assistant to read and analyze the file; it does not import data into the database. For bulk imports, first ask a database administrator to confirm the import approach. Contact the Timecho team for further assistance.

Scanned documents:

• Model: PDFs without a text layer require a conversation model with image understanding to read each page. If the current model does not support this, create a new conversation with a suitable model and upload again.
• Recognition: Wait until Image Recognition in Progress disappears from the attachment bar before sending.
• Pages and usage: By default, only the first 20 pages are recognized. Remaining pages are not read automatically. Each page consumes model usage.
• Interrupted recognition: Upload again if a service restart interrupts recognition.
• Mixed PDFs: Scanned pages in a PDF containing both text and scanned pages may be missed. Extract the required scanned pages into a separate file and upload it.

9. Use the Workspace

Ask "Save this inspection SQL as inspection.sql in the workspace," then "Change inspection.sql to aggregate by hour."

Open Workspace on the right side of the conversation page to see the file list. Click View to read, copy, or download content, or Delete to remove files you no longer need.

Saving does not execute: The workspace stores text. Saving a script does not mean it has run.

• Access and cleanup: Only you can view the files. They are removed when the conversation is permanently deleted.
• Capacity: By default, at most 512 KB per file, 100 files per conversation, and 5 MB in total.
• Insufficient space: Download files you want to keep before deleting older files.

10. Use Images

Paste or drag an image into the input box and ask the assistant to interpret it. The selected model must support image understanding. If it does not, switch as instructed by the interface.

Remove passwords, customer identity details, and other sensitive content before uploading. Images are visible only to the uploader and are removed when the conversation is permanently deleted. Do not upload highly sensitive images that must not be sent to the model service.

11. Save Preferences

Add personal preferences in Memory, such as "Interpret dates using Beijing time by default." You can also extract memories from conversations. Memories affect later responses, so update anything incorrect or outdated.

Do not store passwords, API keys, or other secrets in memory.

12. Manage Conversations and Stop Responses

Task	Action
Continue a discussion	Open the original conversation and ask another question
Rename	Rename in Conversation Management
Handle a long conversation	Generate a summary in Conversation Management or create a new conversation
Stop the current turn	Click Stop Generating
Delete history	Delete in Conversation Management; download important reports first

Continuing or stopping a response: Closing or refreshing the page does not automatically stop generation. Reopen it to continue viewing. For long conversations, the system may summarize older content before a new turn while retaining the original text of recent turns.

After deletion:

• Conversations first enter the trash, with a default retention of 30 days. During that period, they can be restored or permanently deleted.
• After expiration, their messages, images, documents, workspace files, and exports are removed together.
• Approval, audit, and usage records remain after a conversation is deleted.

Note: Deleting a conversation or stopping a response does not undo database operations already submitted.

13. View Usage

Open Usage Statistics to view your usage and daily budget. A token is a unit used to measure model consumption.

Budget exhausted: Ask an administrator to adjust it, or wait until the next day.

Cost interpretation: Amounts marked Estimated are not exact supplier billing figures. A budget is not a strict spending cap.