Understand data management in Insights
It is important to understand how Insights handles your data.
The Insights app helps answer questions by combining information in your organization and your work context. To do that, it can use:
Atlassian apps via the Teamwork Graph for looking up work items, and pages in the apps you have access to, like Jira, Confluence, and Bitbucket.
Atlassian Data Lake, for analytics over Atlassian app data that your site has enrolled in the lake.
Third-party data sources, such as Snowflake, BigQuery, and similar warehouse or BI tools that you connect to via APIs.
These are distinct connections. Connecting to the Teamwork Graph supplements the direct integration with a warehouse, rather than being a replacement or interchangeable.
Analyses
Each Insights conversation is stored in Analyses so you can come back to it later and continue where you left off. This includes your query, the assistant’s response, and any output (such as a markdown table from Atlassian Data Lake or a connected third-party data source).
Insights trims stored history so each completed turn keeps its own slice of the thread rather than duplicating the entire conversation every time. It does not strip query previews out of that turn.
データの使用
To find answers to your queries, you’ll need to connect Insights to internal and external data sources. Once this is done, Insights uses the context from these sources to provide accurate answers. Aside from Atlassian Data Lake, which is connected when you set up Insights, you can connect to other data sources at your discretion.
Here’s how your data connections are utilized:
Atlassian Data Lake
Information from the data lake is used to answer questions related to the apps that your organization has enrolled in (for example, Jira, Bitbucket, and Jira Service Management).
Read about Atlassian Data Lake
チームワーク グラフ
You’ll need to provide permission to connect Teamwork Graph to Insights. When connected, Insights uses Teamwork Graph to look up your work items, pages, and other related work context across your Atlassian apps.
Unlike Atlassian Data Lake, which includes information from lake-enrolled product datasets (aggregated/tabular) and is queried with SQL, Teamwork Graph is a live lookup of a user’s work context within the Atlassian apps they use and connected data sources.
For analytics sources connected through Teamwork Graph (including both internal and external sources), Insights periodically reads catalog information, such as table and column names, types, and descriptions, so people and the assistant can find and understand datasets. Reading the catalog does not involve copying the contents of the tables.
サードパーティのデータ ソース
Once connected to a third-party data source via API integration, such as a warehouse or BI table, Insights will have read-only access to the datasets in the location. When a question needs to be answered, Insights will run a query against the connected source.
How Insights queries a third-party data source
When you ask a question that requires row-level values from a third-party data source:
Insights generates a read-only query.
It checks that you are allowed to run that query on the connected source, including whether you can access the tables named in the query, your team’s access, and any restrictions configured for the assistant.
If the source is set up to use each person’s own credentials, Insights runs the query as you, after confirming your connection. Otherwise, it uses the integration credentials your administrator configured.
It opens a live connection to your source and runs that query there. Results come from the source at the time of the request, not from a stored dump of the table.
A limited number of rows from that query are kept so the answer can be shown, reused in the conversation, and (when the result is large) offered as a download.
Insights adds a short comment to the SQL so your own warehouse logs can show which user and job ran it, except on Salesforce.
What is stored after a SQL query
After a successful run, Insights retains:
The SQL text on a hidden, assistant-generated query record.
A sample of the results from that query (not the full table), in our database. Access to that grid is restricted; it is treated as confidential.
A CSV of that result in object storage, unless upload is skipped. Larger results are linked as a download rather than pasted in full into the chat.
The SQL text in an internal product log used for this query type (kept out of ordinary service logs).
The SQL text in product analytics events used to record whether the authorization check allowed or denied the query.
The preview shown to the assistant in the conversation record, so the thread can be resumed and displayed later.
While the query is in flight, Insights also keeps a short-lived job record (on the order of minutes). The in-flight query text in that job’s working context is encrypted with a workspace-scoped key.
How Insights queries Atlassian Data Lake
When the assistant needs information from Atlassian Data Lake, it uses the same read-only SQL path used to query a third-party source.
The query is a read-only select-style statement.
It is executed live against Atlassian Data Lake at request time, as the signed-in user. Background jobs cannot run ad-hoc lake SQL without that user context.
Access is permissioned. The query is scoped to the lake workspaces and products your organization has enrolled in, and the analytics layer enforces what that user is allowed to see. Insights does not rely on the warehouse-style catalog table-permission check for this source; permissioning is enforced by Atlassian Data Lake and the signed-in user’s identity.
A limited result from that query is returned to the assistant and retained the same way as other Insights SQL (truncated grid, optional download file, conversation preview, query text).
The lake, not Insights, remains the system of record for those product analytics tables. Insights stores only the catalog metadata and the truncated result of queries it ran.
Data previews
When someone opens a table in the catalog, Insights can show a data preview. That preview is loaded live from the connected source at the time of the request and is not a stored copy of the table. Only people with permission to preview that asset can request it. Insights reads a small sample of rows (up to fifty rows; fewer for some document sources), drops any columns marked as PII in the catalog, and returns that sample to the browser. The preview is not saved onto the table as a lasting extract.
A short-lived job record may hold the in-flight sample for a few minutes so the request can complete; that is not a warehouse replica. If the source uses the viewer’s own sign-in, Insights may ask them to connect before showing rows. Charts and dashboards use a separate live preview (for example, a URL or snapshot from the BI tool), not a copy of the underlying dataset.
Google スプレッドシート
If Google Sheets is used as a query source, Insights may keep a short-lived snapshot of a sheet (about 30 minutes) to serve follow-up queries.
この内容はお役に立ちましたか?