Connect Data
Connecting data is the foundation of every later workflow in Tukun.ai. A fast setup is useful only if the source is scoped correctly and can support reviewable analysis.
Where to connect data
Section titled “Where to connect data”You can start from the Workbench or from the data source list.
In the Workbench, open the + menu next to the input:
- choose Upload file for a local Excel or CSV file
- choose Connect database to configure a database connection in the conversation
- choose Select data source to attach an existing source to the current conversation
The database connection flow appears as a new Workbench turn. The assistant reply contains the connection form, so you can enter host, port, database, credentials, and related settings without switching to a separate add-source page. When a connection is created, validate it with a small known question before using it for broader analysis.
The data source list is still useful for source management: reviewing existing sources, opening details, cleaning up obsolete sources, and starting from a full list when the Workbench menu only shows common sources.
What to prepare before connecting
Section titled “What to prepare before connecting”Gather these inputs first:
- a read-only credential whenever possible
- the approved database, schema, or dataset scope
- one or two tables you already trust for validation
- the owner of the source connection
- any sensitive-data restrictions the account should respect
If you do not know which tables are approved for business use, pause setup until that is clear.
Recommended first connection
Section titled “Recommended first connection”Your first source should be:
- representative enough to answer a real question
- narrow enough to validate quickly
- owned by someone who can answer schema questions
Do not start with the biggest possible schema. Start with the smallest source that proves the workflow.
Connection principles
Section titled “Connection principles”Prefer least privilege
Section titled “Prefer least privilege”Use read-only credentials and limit access to approved schemas or datasets.
Prefer service ownership
Section titled “Prefer service ownership”Use a service account instead of a personal user credential whenever possible. This makes maintenance and offboarding safer.
Prefer explicit scope
Section titled “Prefer explicit scope”If a source includes production, sandbox, and test data together, separate them intentionally or document how business users should distinguish them.
The validation step after connecting
Section titled “The validation step after connecting”Do not move directly from “connection succeeded” to “the product is ready.”
Ask a simple validation question against a known table or metric:
Show daily order count for the last 7 days.Check:
- whether the source returns data at all
- whether the row counts look plausible
- whether the time range behaves correctly
- whether the business grain matches expectations
Common onboarding mistakes
Section titled “Common onboarding mistakes”- connecting too many sources before validating one
- using a personal credential that later disappears
- exposing a schema no one is willing to support
- skipping validation because the UI says the connection worked
- mixing production and test data without naming or governance
When a source is ready for broader use
Section titled “When a source is ready for broader use”A source is usually ready for wider usage when:
- one or more representative questions have been validated
- the owner of the source is known
- the allowed scope is understood
- the team knows which metrics still need semantic definitions
Ongoing source hygiene
Section titled “Ongoing source hygiene”After onboarding:
- rotate credentials according to policy
- remove demo or obsolete connections
- rename unclear sources
- revalidate important questions after major schema changes
Source quality is part of answer quality. If the source is weakly owned, the analysis will stay weakly trusted.