A practical reference to Blink's functional boundaries, so you know what to expect and how to work around its limits.
Blink is your AI copilot, connected to the DataGalaxy Catalog through MCP tools. It's designed to help you find, understand, and act on the knowledge already in your catalog. To get reliable results, it helps to know where Blink's functional boundaries are. This article covers what Blink can and can't do operationally. For content and safety restrictions (forbidden topics, prompt-injection protection, content filtering), see Guardrails for the Chatbot.
What Blink can do
- Search the catalog and find objects by name, type, owner, tags, or dates
- Explain an object — its definition, business context, and custom attributes
- Tell you where data is stored (location, hierarchy, full path)
- Trace lineage (how a value is built) and usage (what consumes an object)
- List the columns of a table or view, or the fields of a structure
- Generate SQL queries against catalogued tables and columns
- Add a comment to an object on your behalf
- Work in multiple languages as Blink detects the message from your question or follows your instruction.
Scope
- Blink works on the DataGalaxy Catalog only. It can't answer about modules outside the catalog.
- Blink answers questions about your data and metadata — it doesn't provide step-by-step interface navigation ("click here").
- Blink is not a platform administration tool. It can't manage users, permissions, or configuration.
Creating and changing context
- Blink can't create new catalog objects or delete anything.
- Blink cannot update an object
- It can't create or complete tasks.
- Comments are plain text only — no @-mentions of users or teams.
Counts and analytics
- Blink can't produce catalog-wide statistics or counts. It is not Blink's issue, just LLM are not good with numbers/
- Search returns a relevant sample of results (around ten shown), not an exhaustive total. For full inventories, use analytical functionalities of catalog.
SQL generation
- Blink writes queries using only the columns that are actually catalogued. It never invents column names — if a column you ask for isn't in the catalog, it will tell you.
- The database engine is often not recorded in the catalog. When the dialect is unknown, Blink writes standard ANSI SQL and recommends confirming the exact syntax with the data owner.
- When a query involves personal or sensitive data, and Blink was able to recognise it, Blink may remind you to check access rights and applicable policies first.
Custom attributes and child objects
- Blink can list a table's columns and fields, but it can't fully enumerate other types of child objects — it can only report how many there are.
- Blink can read an object's custom attributes, but it won't necessarily surface every one by default. Ask about a specific object to retrieve its full set of attributes.
Permissions
Every answer respects your access rights. Blink never bypasses permissions or governance rules, and won't suggest workarounds that would. If you can't see an object in the catalog, Blink can't surface it for you either.
Knowledge boundaries
Blink relies entirely on what's documented in the catalog. If information isn't catalogued, Blink can't know it — answer quality depends directly on how complete and up to date your metadata is.
When something goes wrong
If Blink reports a technical error, it won't retry automatically. Wait a moment and try again. If the same issue persists, start a new conversation. If it still continues, go to Profile > Support to create a ticket.
If Blink says an object may have been deleted or is no longer accessible, this can happen when the object was removed after a request was made — search for it again to confirm.
Quick reference
| Capability | Supported? |
|---|---|
| Search, explain, locate, trace lineage and usage | Yes |
| List a table's columns / a structure's fields | Yes |
| Generate SQL on catalogued columns | Yes |
| Add a comment | Yes |
| Update an existing object's attributes (status, owners, tags) | No |
| Create or delete catalog objects | No |
| Create or complete tasks | No |
| @-mentions in comments | No |
| Catalog-wide statistics or counts | No |
| Platform administration / UI navigation | No |
Getting the best results despite these limits
Most limits are easy to work around with clearer prompts: be specific about the object and workspace, ask for one action at a time, and break complex requests into steps. For a full guide, see Best prompting practices for Blink.