AI Support

Support that reads first, and answers only what it can prove.

Boreon Support is the agentic console behind every Tableau Server, Tableau Cloud and TabTotal team. It reads your whole environment, from a Tableau Server’s own service logs to your warehouses, works the problem with you, and opens a case for a person the moment it cannot prove the answer. It reads. It never writes. And it does more than answer: every conversation feeds real-time analytics that surface what your users need while it still matters.

Talk to Boreon Support See what it does

Reads everything

It reads your whole environment, read-only

Server, Cloud and the warehouses beneath them. Boreon Support reads across all of it with a read token, grounds every answer in what it finds, and writes none of it back.

Tableau Server, through TSM. It reads a Tableau Server the way an administrator does: node and process status, and the service logs behind them, from vizqlserver and the backgrounder to the gateway. When a refresh or a service is failing, it reads the log and says what happened in plain words. That is the half a Cloud-only assistant never sees.
Tableau Cloud and your warehouses. On Tableau Cloud it reads your content, your roles and your background jobs. On Snowflake and Databricks it reads the catalog and runs read-only SQL, reaching the table underneath a dashboard whenever an answer needs it.
Grounded in the library and in your own words. The whole of Tableau’s documentation sits within the agent’s reach, alongside your own semantic layer: your metric definitions, your glossary, your field names. Answers use your names and Tableau’s own guidance, and when the context cannot settle a question the agent says so.

Ask, and get an artifact back

The agent does not just describe the problem. It hands you the thing that shows it.

Column-level lineage, drawn. Ask where a number comes from and the agent traces the field from the workbook back through the data source to the warehouse column it reads, and draws the chain. The path is visible, end to end.
A plain-words read of the logs. Point it at a failing extract refresh or a Tableau Server service and it reads the logs for you, names the service that failed, and explains what went wrong in language a person can act on. No log-diving, no guesswork.
A live view for the diagnosis. When a picture settles it faster than a sentence, the agent embeds a live Tableau view in the conversation through a short-lived, view-only token. It shows the problem; it never writes back.

Every artifact is built from what the agent just read, read-only, and it is yours to keep. Nothing here changes a thing on your site.

Agentic Studio

We design what the agents do

The Agentic Studio is our canvas. We wire an agent together the way it would be drawn on a whiteboard, and save it as a Kindness Procedure that runs for one client or for everyone.

Design what the agents do. Drag blocks onto the canvas, wire them together, and save a Kindness Procedure. Kindness, grounding, and the read-only law are always on.

A procedure is a graph of blocks. An email arrives, the sender is matched to your CRM, the sentiment is read, a kind and grounded reply is drafted, and the whole thing waits at a gate until a person approves it. The catalog behind the blocks is forty commands across four families: CI/CD, workflow, the Tableau REST API for Server and Cloud, and TSM commands for Server.

The parts that could change something are held back by construction. Triggers, writes, and TSM commands are human-gated: the agent reads and drafts, and a person authorises any change. There is a global default procedure for every client without their own, and ready-made procedures to start from.

Kindness. The empathetic voice is written deterministically. It never depends on a model being in a good mood. The agent supplies the grounded middle; the warmth is the platform's own.
Grounding. An answer comes only from the account, the thread, and the documentation the agent retrieved. When the context does not settle a question, the agent says so and offers a case rather than guessing.
The read-only law. Nothing the agent touches is written. It reads and diagnoses across your account and, when you connect it, your own Tableau site. A person, or a governed and approved change, makes every actual edit.

Where it sits

Over your Tableau site, beside TabTotal

Boreon Support reads your Tableau Server or Cloud from above, read-only. For a TabTotal customer it runs in parallel with TabTotal over the same site, and neither one gets in the other’s way.

It reads, read-only. Support connects to your Tableau Server or Cloud with a read token and reads it: your content, your roles, and which extract refreshes are failing. It never writes.
It runs in parallel. For a TabTotal customer, Support sits alongside TabTotal over the same site. One is your support platform and the other your governance platform, and they share the reading.
One site of record. Both work over a single Tableau Cloud or Server site. The picture an agent grounds an answer in is the picture your team already trusts.

It reads. It never writes. Every actual change stays with a person, or with a governed and approved step.

An agent that works the problem, then finds a person

The customer console opens on one question: how can Boreon Support help? What follows is a real conversation, grounded in what the agent can see.

It reads your account, read-only

The agent sees your account, your licence and seats, and your open cases. Connect your own Tableau Server or Cloud site read-only, with a personal access token or a Connected App, and it reads that too: your content, your roles, and which extract refreshes are failing.

It opens a case that carries everything

When a person is needed, one button turns the whole conversation into a case for the team. Nothing is retyped, and nothing is lost between you and the engineer who picks it up.

Honest about its limits

A safety screen runs before any model is called, and fails closed. If the agent cannot prove an answer from what it has, it does not invent one. That restraint is the reason the answers are worth trusting.

Real-time conversational analytics

Every conversation is more than a ticket. As they happen, the platform reads the themes, the sentiment and the volume across them, and you see what your users are struggling with while you can still act on it. The reading is grounded in the conversations themselves, and read-only like everything else.

More than support

TabTotal tools, without a TabTotal licence

Boreon Support brings some of TabTotal’s most useful tools to any Tableau Server or Cloud customer, with no TabTotal licence needed. Each one reads your site read-only and leaves the result in your hands.

PII and PHI risk scanning

Scan your Tableau Server or Cloud for exposed PII and PHI, the same scanning that powers TabTotal Risk. It reads your content and reports what it finds, and it changes nothing.

Query your data, build a dashboard

Ask a question of your data and get a live React dashboard back, a lite edition of the Build studio. Full querying and a dashboard studio, inside the conversation.

Tableau Cloud backups

Scheduled, read-only backups of your Tableau Cloud content to your own S3 bucket or Azure Blob storage. The copy lands in your account, never in ours.

Analytics on your dashboards

Ask your Tableau dashboards questions in plain language and read grounded answers. The reading is read-only, the same as everything else here.

It meets your team where they already work

Email, chat, your warehouse metadata and a live dashboard, each connected in the way that keeps it read-only.

It meets your team where they already work
	What it is	How it stays read-only
Agentic Inbox	Mail to support@boreon.com becomes a conversation. Each message is matched to a contact, tied into that account's thread, and answered by the Email Agent, the kind heart of the inbox.	The agent drafts; a person sends. Empathy is deterministic; a warm reply never depends on the model.
Microsoft Teams	New email and new cases arrive as cards in the channel your team already watches. Support stops living in a tab nobody has open.	A one-way notification into a channel. It tells your team something happened; it changes nothing on its own.
MCP relay	A read-only bridge to the TabTotal master gateway. The agent lists projects, workbooks, data sources, views and metrics to ground an answer in your own environment.	The relay refuses write verbs at the protocol layer. The bridge can only ever read, whatever the question.
Connected App (JWT)	A live dashboard embeds directly in the conversation for diagnostics, and the agent reads the shape of a site: users and roles, data sources, and extract-refresh health.	Short-lived tokens scoped to viewing. The embed shows; it never writes back, and no data leaves your grounded answer.

Hybrid AI

Two model makers, the right one for each task

Boreon Support is mixed by design. Every task is routed to the model that handles it best, drawing on the strengths of both OpenAI and Anthropic. The mix is deliberate, and it is how the platform stays fast where speed matters and careful where the stakes are high.

OpenAI. The GPT models classify each message, draft the replies, and power retrieval across your account and the documentation. This is the fast, high-volume work of a live conversation.
Anthropic. Claude reasons through the escalations, the moment a case is handed up and the thinking has to be careful. The harder the question, the heavier the model it earns.

A deterministic safety screen runs before either maker's model is called, and the read-only law holds whichever one answers. The hybrid changes which model thinks; it never changes what the platform is allowed to do.

What is always on

The guarantees do not depend on which procedure is running, or on whether an AI key is configured at all.

Read-only, by construction

The Tableau client has no method that writes, publishes, refreshes or deletes. The boundary is the code, where no setting can move it.

Grounded or silent

Answers come from retrieved context. No usable context, no invented answer, and the model is not even called.

Deterministic kindness

The caring tone is written into the platform. A reply reads the same whether or not a model is having a good day.

A safety screen first

A deterministic screen runs before any model call and fails safe, with crisis resources where they matter and every block audited.

Bring us the problem. We will read it first.

Support is included in the TabTotal Enterprise License for all TabTotal customers. Tell us what is happening with Tableau or TabTotal, and an agent starts working it before a person ever has to.

Talk to Boreon Support