01 · Understand
“Summarize this dashboard.”
Key statements, a metrics table, the active filters. Every figure with the worksheet it came from.
OpenVizPilot for Tableau
Ask your dashboard.
OpenVizPilot answers questions right inside the Tableau dashboard, using the current context and the permissions of the signed-in user.
- Self-hosted
- BYO LLM
- Tableau session
- Row-level security
- v1.7.8
Executive overview – profitability · Tableau
Executive Overview - Profitability (All)
Sales
€2,326,534.35
Profit
€292,296.81
Profit ratio
12.6%
Profit per order
€57.28
Monthly sales by product category
Office Supplies
Furniture
Technology
OpenVizPilot
- I know this dashboard – its figures, filters and worksheets. Pick a question:
- What stands out in the data? Name the three most important points.
-
✓ aggregate_summary_data · “Sales by product”
worksheet: "Sales by product" groupBy: ["Category"] measures: ["SUM(Sales)", "SUM(Profit)"]
→ Grouped across all data pages · no full-data permission required
-
The three most important findings
-
1. Low overall profit ratio despite solid revenue
On total revenue of €2,326,534.35 the profit ratio is only 13% (profit: €292,296.81).
-
2. Furniture is the weakest category
Only 14.4% margin on profitable orders – well below Technology (~25.8%) and Office Supplies (~28.6%). It also carries the largest unprofitable share at €233,177.
-
3. Office Supplies has the worst loss ratio on unprofitable orders
A loss of -€50,525 on €113,808 of revenue – roughly -44%.
Tip: click a source to highlight its worksheet in the dashboard.
-
- Let me know when the profit ratio drops below 10%.
- ✓ propose_watch_rule · “Total sales”
- I’ll set that up as a watch rule. Review it – nothing is created until you confirm.
-
Create this rule?
- Name
- Profit ratio below 10%
- View
- Total sales
- Threshold
- below 10%
- Schedule
- Daily
- Channel
- Microsoft Teams
- Target (URL or email address)
- https://company.webhook.office.com/…
This rule reads data server-side on your behalf — like other server queries, this requires your consent. The rule keeps running even when you are not signed in.
- ✓ Rule created. OpenVizPilot checks daily and alerts you when the value drops below the threshold, then at most every 24 hours – and sends an all-clear once it recovers.
-
How an alert arrives in Microsoft Teams (example)
Microsoft Teams · OpenVizPilot
Alert: Profit ratio below 10%
- View
- Total sales
- Value
- 9.8%
- Threshold
- below 10%
- Time
- 2026-10-05, 08:00
- Rule by
- anna.berger
The profit ratio has dropped below the 10% threshold.
Open analysis - Discarded – nothing was created.
- The answers in this demo are prepared. Pick one of the sample questions – in real use, OpenVizPilot answers any question about the dashboard.
Click a question: sources highlight their worksheet, tool calls expand, and a rule is only created after you confirm.
Illustration with sample data. The answers come from real product output but run here without AI.
Three questions dashboards can’t answer. OpenVizPilot can.
02 · Investigate
“What stands out in the data?”
OpenVizPilot aggregates the summary data itself and names outliers, weak categories and unusual ratios. With figures, not impressions.
03 · Act Enterprise
“Show me only the last 90 days.”
The filter arrives as a plain-text proposal. You click, the dashboard changes. Not the other way round.
Investigate mode
“Any idea why it dropped?”
The assistant first writes an analysis plan, works through it with several tool calls, and closes with main cause, evidence and sources.
Your permissions stay your permissions.
OpenVizPilot reads dashboard data in the Tableau session of the signed-in user. Row-level security and user filters apply unchanged. The middleware has no Tableau identity of its own – no service account, no personal access token. It caches no dashboard data and stores no chat history.
-
Extensions API
Data access runs in the user’s browser, in their Tableau session.
-
Summary data
No full-data permission needed. The extension reads only what the dashboard shows.
-
Click, not automation
Filters, parameters and highlights change only when the user clicks a chip.
-
Scope guard
A small classifier model rejects off-topic questions server-side before they reach the main model.
Your infrastructure. Your LLM.
The extension and middleware run on your side: with Docker Compose on a Linux VM or via Helm chart on Kubernetes. You connect the language model through an OpenAI-compatible endpoint – for example LiteLLM, vLLM, Ollama or a proxy in front of Azure OpenAI. Where questions and data context go is your decision.
- Docker Compose
- Helm / Kubernetes
- OpenAI-compatible endpoint
- Secrets from your vault
- One origin, no CORS
What Enterprise adds.
Same extension, same chat window. Enterprise adds what companies need for a rollout.
-
SSO
Single sign-on
OIDC with Microsoft Entra ID, Keycloak or other providers. The ID token is verified on every request.
-
Actions
Dashboard actions
Filters, parameters, highlights, showing and hiding zones – proposed as a chip, executed on click.
-
Memory
User memory
Role and preferred views are remembered. Visible and deletable in the panel.
-
Saved queries
Saved questions
Answer focus and up to five standard questions per dashboard and user.
-
MCP
MCP sources
Approved data sources and web search, assigned per site, dashboard and member. External queries only after consent.
-
Tableau Server
Tableau Server integration
Workbooks, views, field definitions and formulas in the same chat. Tableau Server 2024.2 or newer, via Connected App.
-
Server Data
Server-side data access
Reads views outside the open dashboard server-side on the user’s behalf – only after admin approval, the site switch and one-time consent. Every access is logged 90 days, pseudonymized, in the audit trail.
-
Watch
Watch (scheduled alerts)
Rules start as a chat proposal confirmed by a human. Checked on a schedule, delivered via webhook, Microsoft Teams or email – the alert text is deleted right after delivery.
-
Cross-Dashboard
Estate-wide investigation
Investigate mode can widen scope to the entire Tableau environment: the assistant also searches published data sources with their fields, tables and dependent workbooks, reads at most five views per question and – behind its own site switch – queries data sources itself (Tableau Cloud or Server 2025.1+). A fixed query budget caps server access, and every answer carries its source list.
Model catalog, slash commands, standard questions and the maximum tool rounds per dashboard, license and SSO are managed in the admin UI – no redeploy. The overview shows the version with an update notice and LLM usage. Core and Enterprise.
How it fits together.
The extension sits inside the Tableau dashboard and executes tool calls in the browser. The stateless middleware adds the system prompt and tool definitions and streams the answer. The LLM endpoint is yours. Enterprise adds further data paths: Tableau Server metadata under the user’s verified identity, data from further views read server-side on their behalf, and scheduled Watch rules.
One license. One production Tableau environment. No per-user licensing.
OpenVizPilot Enterprise
€14,500
net / year per production Tableau environment
Associated dev, test and staging environments included.
No additional per-user license.
AI model and infrastructure costs are borne by the customer. Planned features are not a contractual commitment.
- Commercial use
- All Enterprise capabilities
- Updates
- Support straight from the developer
- Self-hosted
- BYO LLM
Smaller Tableau installation?
We offer adjusted terms for smaller teams. Tell us briefly how many dashboards and users are involved.
Launch package
€1,500 one-time
Kickoff and architecture review, installation and production deployment, SSO via Entra ID or Keycloak, connection of an OpenAI-compatible LLM endpoint, integration in up to three dashboards, validation of permissions and row-level security, admin training, up to 8 hours of go-live support.
Additional dashboards, integrations or customizations are quoted by effort.
Core or Enterprise?
Core answers questions about the open dashboard. Enterprise adds company sign-in, personal assistance, confirmed actions and additional data sources. It is the same extension with the same chat window.
Core
For analysis of what users can see in their dashboard. Admins manage models and shared analysis prompts; local accounts provide sign-in. The source is available under PolyForm Noncommercial (source available) and free for noncommercial use. Commercial Core use also requires an agreement.
Enterprise
For teams needing company sign-in, personal settings and additional data sources. Personal saved questions belong to the individual; shared dashboard questions are already in Core. Extended discovery of Tableau content and field definitions belongs to Enterprise and uses the same chat window.
The matrix maps capabilities to editions, including intended additions; it does not imply immediate availability of every feature. The binding scope and availability for your deployment are agreed in the offer.
| Capability | Core | Enterprise |
|---|---|---|
| Dashboard chat: visible measures, filters and parameters | Assigned | Assigned |
| Source references, visible tool calls and Markdown export | Assigned | Assigned |
| Your model endpoint and a centrally managed model catalog | Assigned | Assigned |
| Glossary, slash commands and shared dashboard questions | Assigned | Assigned |
| Local user accounts and admin interface | Assigned | Assigned |
| Self-hosting via Docker Compose or Helm; aggregate usage overview | Assigned | Assigned |
| Approve AI chat and Tableau API access separately per user | Assigned | Assigned |
| SSO via Entra ID, Keycloak or other OIDC providers | Not assigned | Assigned |
| Personal memory, answer focus and saved questions | Not assigned | Assigned |
| Dashboard actions: filters, parameters, marks and zones after a click | Not assigned | Assigned |
| Approved MCP sources and web search with confirmation | Not assigned | Assigned |
| Tableau Server: find accessible workbooks and views in chat | Not assigned | Assigned |
| Explain field definitions and calculation formulas | Not assigned | Assigned |
| Data dictionary: fields, descriptions and known dependencies | Not assigned | Assigned |
| Visually trace data origins and the impact of changes | Not assigned | Assigned |
| Explain data freshness, datasource health, access rights and changes | Not assigned | Assigned |
| AI assistance for administration and server-wide analysis | Not assigned | Assigned |
Checkmark = assigned to this edition. Enterprise also covers Core capabilities. A user approval in Core does not replace an Enterprise license.
One chat. Two questions. Two data paths.
Ask both questions in the OpenVizPilot chat inside your dashboard. Only the queried source differs.
“Which region is below target?”
Core + Enterprise
Chat → question → visible dashboard data in your Tableau session → AI analysis → answer with sources in the same chat.
“How is margin defined?”
Enterprise
Chat → question → accessible field definitions and formulas via Tableau Server APIs → AI explanation → answer with sources in the same chat. No second chat window.
Your IT team sets up the additional Tableau connection. Users sign in with their company account and see only content they are permitted to access. Admins approve AI chat and Tableau API access separately; API questions in chat require both. The extension explains available metadata and formulas, not arbitrary data across the entire server. Answers can be checked against their sources.
Source available
The OpenVizPilot core is published under the PolyForm Noncommercial license and free for noncommercial use. Commercial use requires an agreement. Enterprise capabilities are proprietary and unlocked with a license key. The code can be audited – before you buy.
Frequently asked questions
- What is OpenVizPilot?
- An AI copilot for Tableau: a dashboard extension with a chat that answers questions about the open dashboard, plus a middleware that connects your LLM. It is made by WerkWorks.
- Where does OpenVizPilot run?
- Right inside the Tableau dashboard, as a dashboard extension in Tableau Desktop and Tableau Server. Tableau Cloud uses the same Extensions API; we align deployment there with you. The middleware runs self-hosted on your side.
- How does OpenVizPilot access data?
- Through the Tableau Extensions API in the session of the signed-in user. It sees only what the user sees.
- Do Tableau permissions and row-level security still apply?
- Yes. Every data access runs in the user’s Tableau session; the middleware has no Tableau identity of its own.
- Can I use my own LLM?
- Yes. Any OpenAI-compatible endpoint – for example LiteLLM, vLLM, Ollama or a proxy such as LiteLLM in front of Azure OpenAI. Model choice and hosting are yours.
- Is Tableau data stored in an external cloud?
- Not by OpenVizPilot. Dashboard data goes as context to the LLM endpoint you configure – where that runs is your decision. The middleware stores neither dashboard data nor chat history. Licensed installations check in with WerkWorks once a day; that activates the installation and renews a seven-day lease. If WerkWorks is unreachable, the installation keeps running unchanged for up to 30 days after the lease expires (grace period), after which only Core functions remain until the next successful check-in. What is transmitted: installation ID, license key, version and two aggregate numbers (active users, dashboards) – never content, names or dashboard data.
- What does OpenVizPilot Enterprise cost?
- €14,500 net per year per production Tableau environment. Dev, test and staging environments of the same Tableau landscape are included. There is no per-user license.
- What about smaller Tableau installations?
- Adjusted terms are available on request for smaller teams.
- Is OpenVizPilot open source?
- Source available: the core is published under PolyForm Noncommercial and free for noncommercial use. Enterprise capabilities are proprietary.
- Which Tableau version is required?
- The extension runs as a dashboard extension in Tableau Desktop and Tableau Server. User memory and saved questions need Tableau 2023.2 or newer; the Enterprise integration with Tableau Server APIs requires Tableau Server 2024.2 or newer.
- Does OpenVizPilot know our own metric definitions?
- Yes. Besides the per-workbook glossary, the admin can maintain a company-wide metric catalog with definition, synonyms and source. When a question matches a cataloged metric, the assistant looks it up and marks the answer with a “✓ Verified definition” badge.
- What does Investigate mode do?
- The assistant first writes an analysis plan, works through it with several tool calls, and closes with main cause, evidence and sources. With an Enterprise license, scope can widen to the entire Tableau environment; the assistant then also searches published data sources, reads at most five views per question and can query data sources itself – bounded by a fixed query budget.
- Is there a ready-made report for leadership?
- Yes, via the /exec-brief slash command (or /fuehrungsbericht): key takeaways, the most important metrics with change, notable deviations, up to three recommendations and the data basis – exportable as Markdown.
- Can I trace where a number comes from?
- With /lineage (or /herkunft): dashboard → worksheet → field → catalog definition → data source → upstream tables → active filters → certification status. Whatever can’t be determined is marked “not available” instead of guessed.
See OpenVizPilot in action.
Request a demo. Tell us how you use Tableau and which questions your team wants to answer in its dashboards.