This release strengthens two parts of the TimeXtender AI offering: the MCP Server that securely connects TimeXtender to your data, and Xpilot Analytics, the conversational analytics experience built on top of it. The MCP Server work makes secure on-premise deployment work end-to-end and returns more trustworthy query results. The Xpilot Analytics work makes the chat experience accurate, readable, and correctly isolated per workspace.
TimeXtender MCP Server
In short
This update makes secure (HTTPS) deployment work end-to-end, returns trustworthy query results, and keeps the service running through reboots. Previously, enabling HTTPS was blocked by a series of separate problems; an administrator can now stand the server up over HTTPS, on a standard port, with the cloud relay, and change or revert that setup later, entirely from the configurator without editing files by hand.
Before and after
| Area | Before | In 2.0.1 |
| Saving an HTTPS setup | Saving an HTTPS configuration could leave the service unable to start, with a misleading "Service not found." | The configuration saves correctly and the service starts; startup errors report the real cause. |
| Running HTTPS as a service account | The service crashed on secure startup with "Access denied" because the certificate was not reachable by the service account. | HTTPS starts cleanly under a real service account. |
| Using a standard port (443) | Standard ports were rejected, so clean addresses like https://host/mcp were not possible. | Standard-port HTTPS works, with no port number in the URL. |
| HTTPS together with the cloud relay | Enabling both caused every relayed request to fail ("no models available"). | Secure HTTPS and the cloud relay work together. |
| Changing HTTPS settings later | Reopening the configurator on an HTTPS server blocked further edits without re-validating the certificate. | A previously validated setup reopens ready to edit, so changes save without redoing validation. |
| Reverting to HTTP | There was no supported way to remove a certificate or switch back to HTTP from the interface. | Switching back to HTTP cleanly removes the secure settings. |
| Service recovery after a reboot | After a reboot or OS update, the service stayed down until someone restarted it manually. | The service starts itself after a reboot and recovers from an isolated startup failure on its own. |
| Calculated measure results | A calculated measure could draw from the wrong table and return incorrect numbers. | Measures are scoped correctly and return trustworthy results, with clearer errors. |
| Multi-field table relationships | When tables were linked on more than one field, only the first was used, silently skewing totals and counts. | Every field in the relationship is honored, so totals and counts match the source data. |
| Names matching database keywords | A field or metric named like a reserved database word made a query fail with a misleading "table does not exist" error. | These everyday names now work normally. |
| Record-count questions | Legitimate "how many records" questions were sometimes blocked as a suspected security risk. | These common questions return real answers. |
| Failure reporting to applications | Failed queries could be returned looking successful, so connected applications could not detect the failure. | Failures are reported as errors, so connected applications detect and handle them reliably. |
| Data model details | The "Schema" field for each table was always blank. | Returns a consistent, meaningful value. |
| Installer package | The installer included a number of obsolete and internal files. | A leaner, more professional installed file set. |
What this means
- Secure deployment is realistic. On-premise HTTPS, standard ports, and the cloud relay work together, set up and maintained from the configurator without hand-editing files.
- Numbers you can trust. Calculated measures, multi-field relationships, and reserved-word names all return correct results.
- Honest failures. When a query fails, connected applications know it failed.
- Less downtime. The service recovers itself after a reboot or transient startup failure.
To upgrade to MCP Server 2.0.1 you can download and run a new installer.
Xpilot Analytics
In short
This update makes the Xpilot Analytics chat experience accurate, readable, and correctly isolated. The assistant now answers honestly when something cannot be resolved instead of quietly substituting a different answer, charts and tables present complete and correct data, the interface is polished (including dark mode), and a workspace-isolation gap is closed so a data source is only usable in the workspace it was assigned to.
Before and after
| Area | Before | In 26.2.1 |
| Honest answers when a query fails | The assistant could quietly swap in a different column or count and present a real-looking result for a question the user never asked. | The assistant surfaces the failure, offers alternatives, and does not present a substitute result until the user confirms. |
| Correct first answer on a new chat | The first question in a conversation often failed because the assistant guessed table and column names. | The assistant checks the real data structure first, so the first question returns the correct answer. |
| Tables and CSV exports | For some data sources, table and CSV exports showed blank cells for totals and counts. | Tables and CSV exports show complete, correct values. |
| Charts that match what was asked | The assistant offered visual changes the chart could not actually make, returning an unchanged chart with no explanation. | The assistant only offers visual options the charts can deliver, and says so when one is not supported. |
| Readable chart labels | Charts combining two values showed unreadable shorthand like "1(1)" or "5(2)". | Labels are shown in full, with a dedicated grouped bar chart for two-dimensional data. |
| KPI cards | Large values overflowed the card, a stray line appeared under the number, and labels were forced into ALL CAPS. | Large numbers are abbreviated cleanly (e.g. 27.2M, 1.4B), the stray line is gone, and labels keep their original capitalization. |
| Code blocks in dark mode | Code shown in chat was nearly invisible against the dark background. | Code has proper contrast and is clearly legible in dark mode. |
| New data models appearing | A newly added data model stayed invisible in existing conversations until a brand-new chat was started. | Opening or refreshing an existing conversation picks up newly added models. |
| Conversation history (data source) | Past conversations showed "Unavailable" instead of the data source they used. | Each past conversation shows the data source it actually used, including ones since removed. |
| Refining a chart | Refining a chart felt like a reply to a separate question, and it was unclear whether the chart had changed. | Refinements are treated as edits to the chart in view, with a clear signal that it was updated. |
| Clarity while the assistant works | The progress panel showed a long, technical tool description on every step. | The panel shows short, friendly action labels such as "Querying the semantic model." |
| Workspace isolation of data sources | A data source not assigned to a workspace could still appear and be used in that workspace's chat. | A data source is only visible and usable in the workspace it was assigned to. |
What this means
- Answers you can trust. The assistant no longer answers a slightly different question without making that clear, and it asks for direction when something cannot be resolved.
- A polished, complete experience. Charts, tables, KPI cards, dark mode, and conversation history all present correct, readable information.
- Tighter isolation. A data source is confined to the workspace it was assigned to, closing a cross-workspace exposure gap.
If your service is on version 26.2 you already have access to these new features.
