Release notes

For a detailed list of all changes, see the Changelog. For details about versioning, see our Versioning Policy.

6.2.1

2026-09-03

The Atoti MCP Server can now connect to other MCP servers, typically other Atoti Server instances in the same deployment. It can then use their tools in chat and through its own MCP endpoint. This requires the Atoti Intelligence Extension tier.
Applications can now turn off chat entirely with a single configuration property. Atoti UI then reports chat as absent. This differs from running without an LLM, where the chat endpoint stays advertised but every call fails. See Configure Visualize This.
The chat assistant can now display the interactive Auto-Explain page for an analysis it just ran, directly in the conversation. This requires Atoti UI 5.2.28 or higher.
Chat and Auto-Explain now each have a dedicated REST namespace, so the two features can be updated and versioned independently. See the migration notes.
Administrators can now set limits on how long a chat prompt may run, and on retries, repeated tool calls, tool errors, and rounds with no tool call. These limits keep AI runs bounded and predictable, stopping a model that would otherwise loop without making progress. See Retries and timeouts.

Dynamic tool discovery, introduced in 6.2.0, can now be turned off, sending the full tool set to the model on every request instead.
A completed Auto-Explain analysis can now be listed, displayed again, or removed from the chat, instead of being run a second time.
Closing or canceling a chat request now stops the underlying work immediately, instead of letting it keep running unread.
A license missing the required AI components is now reported clearly at startup, instead of surfacing later as a generic error in Atoti UI.
The MCP server now stays disabled when the license lacks the required tier, even if the application tries to enable it through configuration.
Chat responses streamed through a reverse proxy no longer stall. New response headers stop the proxy from buffering or compressing the event stream.

2026-07-27

The Atoti chat now uses dynamic tool discovery. Instead of sending every available tool with each request, the model searches a keyword index for matches. This reduces token usage and improves tool selection.
The Atoti Java SDK now includes a Dimension API to control which cube attributes are exposed to the AI tools.
Auto-Explain, which identifies the root causes behind measure variations, now has a Python API in the Atoti Python SDK. See How to set up Auto-Explain in Python.
The Atoti Python SDK now offers Session.chat to start a chat session against a cube.
The Atoti Python SDK now exposes the AiConfig.disclaimer attribute, letting the disclaimer shown with AI responses be customized. See How to set up a custom disclaimer in Python.

The chat system prompt was improved to help the LLM navigate the UI.
Fixed the Atoti MCP Server credentials page, which failed to load when the application ran under a custom context path.