18 things teams actually do with AI on their own data.

Data and analytics teams

Ask your data, in your words

A business question lands and the person who knows the schema is in a meeting. The answer waits for a ticket, a view, or another spreadsheet.

Spark turns the question into governed SQL on the live source or target, runs it read-only under the caller's permissions, and returns the numbers with the SQL beside them. The same question also works as a gluesync.ai row from any JDBC client.

Try

How many orders did we sync yesterday, broken down by country?

BI and reporting teams

Let the BI tool ask the question

Tableau, DataGrip, DBeaver, and the reporting job from 2019 all speak SQL. The AI stack speaks HTTP. Someone has to write the adapter nobody wants to own.

The gluesync.ai virtual table is served by Query Forge like any other table. A SELECT with a question returns one row with the plain-language answer, a proposed SQL draft that is never auto-executed, citations, and a conversation ID for the follow-up. Read-only tools only, the caller's token only.

Try

SELECT answer, proposed_sql FROM gluesync.ai WHERE question = 'which customers churned last quarter in the crm-sync pipeline?';

Analysts and team leads

Describe a chart and keep it current

Visualize ships with Gluesync 2.3.1

A simple chart of a synced entity takes a BI license, a dataset, and a refresh job before anyone sees a bar.

Describe the chart to Spark. Visualize grounds it to real columns through the Enterprise brain, renders it in the Control Plane, refreshes it by hand, on a Chronos schedule, or when the bound entity changes, and publishes a chart or a whole canvas at a stable public URL.

Try

Chart monthly order totals for the last year from the ORDERS entity and refresh it every night.

Enterprises with mainframe and midrange estates

Bring IBM i, Db2, and Oracle into AI without touching the host

The most valuable data sits in systems that no AI vendor connects to, and the teams that own them will not install anything new on the box.

Gluesync already replicates from IBM i, Db2, Oracle, SQL Server, and 16 dialect engines. Spark, Query Studio AI, and gluesync.ai reason over the schema Gluesync discovers and query the replica or the source read-only, so legacy data joins the AI estate without a new agent on the host.

Try

Summarize open invoices older than 90 days from the IBM i ledger and propose the SQL you used.

Operations and DBAs

Triage sync health before anyone asks

Pipeline status, per-entity progress, agent metrics, and webhook dead letters live in four screens. Correlating them during an incident costs the minutes you do not have.

One question returns pipelines in error, the entity and reason behind each, growing source lag, and the dead letters that explain a silent webhook, correlated in a single answer.

Try

Which pipelines are in error right now, and why?

On-call engineers

Explain this alert

A CRITICAL notification arrives at 02:00 with an entity name and an error code. The runbook is three clicks and one wiki away.

Ask AI Studio to explain the notification. Spark receives a read-only tool set scoped to that alert's pipeline, agent, and entity, searches the documentation and anonymized support knowledge, and answers in plain language with no write path disguised as a summary.

Try

Explain this alert and tell me whether it needs action tonight.

Operations leads

Put the answer on a schedule

The same health question is asked every morning. Somebody opens the dashboard, copies numbers into a message, and forgets on Fridays.

Any answer Spark can produce once, it can produce every morning. A Chronos schedule fires a published agent version, the prompt template fills in the context, and the digest is delivered by email or webhook before the stand-up.

Try

Every weekday at 8am, email me a report of pipelines that fell behind overnight.

Operations leads

Reach the right people, the right way

An agent found the problem. Now it has to tell someone, with the right severity, through channels security already approved.

Spark escalates through the webhooks and alert emails Core Hub already has configured, at the INFO, WARNING, or CRITICAL severity you ask for, and can summarize and mark the day's notifications as read.

Try

Notify ops by webhook and email that the orders pipeline is stalled.

Platform engineers

Improve a pipeline configuration

Write settings, polling intervals, and column mappings were chosen on day one and never revisited. Changing them by hand means reading YAML nobody remembers.

Spark reads the current configuration, explains what it would change and why, and applies it only after a Manager or Super admin confirms. Nothing executes on the first call; a pending action expires after five minutes.

Try

Review the orders-to-analytics pipeline and suggest safer write settings.

Platform engineers

Stand up pipelines from a conversation

Three new pipelines mean three trips through the wizard, each with the same connection details and the same mapping decisions.

Describe the pipelines once. The Pipeline creator keeps an ordered multi-pipeline plan in the conversation, resumes the first incomplete item on continue, scrubs credentials before they reach the model, and runs the same completion transition the wizard uses so the pipelines leave Draft when they are actually ready.

Try

Create three pipelines from the SAP source to PostgreSQL analytics: customers, orders, and invoices.

Integration engineers

Automate inside and outside Core Hub

Your ETL needs a fresh snapshot of one entity on demand, and your release process needs a pipeline drained and restarted without a person watching the screen.

Trigger flows with their own token, sync commands, and webhooks let Spark run the operation and hand the result to your own systems. Destructive tools park in the chat with Confirm or Cancel; writes from API keys wait for a signed-in approval.

Try

Create a trigger flow that runs a one-time snapshot of the invoices entity on demand.

Data quality teams

Validate a migration and reconcile the differences

After a snapshot, someone still has to prove that source and target agree, and explain the rows that do not.

Spark starts a Validator comparison, reads the result, and explains the differences in business terms. In a chained Chronos event the comparison runs right after the snapshot, and the agent's summary reaches the team before the morning check.

Try

Compare CUSTOMERS on source and target after last night's snapshot and summarize the differences.

Developers

Embed a governed agent in your own application

Product wants an assistant inside the customer portal. Security wants to know which model it calls, with whose key, and what it costs.

Your application starts runs through the native /api/ai/v1 contract or the OpenAI-compatible /v1/chat/completions endpoint with a scoped gsa_ key. Budgets return 402 before spend runs away, rate limits and allow-lists bound the key, and every run keeps a redacted outcome and an ordered event timeline.

Try

POST /api/ai/v1/agents/support-triage/runs with an idempotency key and a correlation ID.

Data governance

Discover where your data hides

Nobody is sure which of the 1,400 tables in the ERP hold personal data, or which ones changed since the last replication review.

Point Spark at a connection and let it walk schemas, tables, and columns, classify PII by column name and optional sample scan, and shortlist the entities worth replicating. Classification returns labels only, never raw cells.

Try

Which tables in the ERP source hold customer personal data?

Security and compliance

Hunt for sensitive columns without SELECT *

Compliance wants proof that every column holding tax IDs, card numbers, or health data is known and masked. Sampling production tables by hand is the risk you are trying to avoid.

classify_table and classify_schema return PII labels only. Query Studio and the AI helper mask the classified cells in the grid, copies, exports, and model prompts, and a per-user data-access matrix decides who may unmask a column. A monthly schedule can run the sweep and report new findings.

Try

List every column classified as PII across the billing pipeline that is not yet masked on the target.

Engineering and product teams

Close the loop in GitLab, Notion, or Slack

The agent found the failing entity. The fix lives in a GitLab issue, the runbook in Notion, and the team in Slack.

Approved company tools are HTTPS MCP servers registered on the Enterprise brain with an allow-listed host and an encrypted token. Agents that carry the tool bundle can open the issue or post the update during a run; remote calls count as writes and wait for a signed-in approval.

Try

Open a GitLab issue for the CUSTOMER_ADDR source timeout and link today's run.

Executives and business teams

Answer the executive question in the chat they already use

Leadership asks faster than the BI team can publish views, and each answer becomes another extract that drifts from the source.

Claude, ChatGPT, Grok, Meta Muse, Cursor, or any MCP client connects to Core Hub, searches the Enterprise brain for the tables behind the question, and hands it to a published Spark agent. Every call runs with the permissions of the person asking; the brain stays on your infrastructure.

Try

Which tables feed the quarterly revenue forecast, and when were they last synced?

Everyone who operates Gluesync

Learn every Core Hub corner

Documentation, support knowledge, global configuration, and license details are each one menu away, and the question is never about just one of them.

Spark answers in context: how Gluesync handles a schema change, which global settings differ from the defaults on this Core Hub, what the license allows, and which past support cases looked like this one.

Try

Which global settings differ from the defaults on this Core Hub?