Changelog

​

Added

• atoti.Level.description ^PYTHON-951.
• atoti.Ordering and atoti.StringComparison ^PYTHON-959. Set atoti.StringComparison.case to "insensitive" to compare letters regardless of their case.
• atoti.Cube.restrictions accepts conditions returned by atoti.Hierarchy.isin() ^PYTHON-895.
• atoti.agg.max(provider_compatibility) and atoti.agg.min(provider_compatibility) ^PYTHON-778.
• atoti_ai.AiConfig.mcp ^PIVOT-14244.
• atoti_ai.AiConfig.chat_enabled ^PIVOT-14244.
• atoti_ai.AiConfig.dynamic_tool_discovery ^PIVOT-14439.
• atoti_ai.AiConfig.auto_explain ^PIVOT-14121.
• atoti_ai.AiConfig.prompt and atoti_ai.AiConfig.max_attempts ^PIVOT-14839.

​

Changed

• Upgraded Atoti UI to 5.2.27.
• atoti.KerberosConfig.username_case_conversion and atoti.LdapConfig.username_case_conversion are applied once, at authentication time, and the resulting username is used everywhere: individual_roles lookup, atoti.Session.user, etc. ^PYTHON-974. With Kerberos, the domain is also dropped: alex@EXAMPLE.COM and alex are the same user, named alex, so individual roles must be assigned to the short name.

​

Deprecated

• atoti.Level.order, CustomOrder, LexicographicalOrder, and NaturalOrder ^PYTHON-959. Use atoti.Level.ordering instead.

​

Fixed

• A data load whose transaction failed at commit time raised a NullPointerException instead of a descriptive error ^PYTHON-778.
• Joining a distributed cluster rebuilt the cube, re-feeding all aggregated data and re-issuing every aggregation query against the database, which can be very expensive in a DirectQuery setup ^PIVOT-14807.
• Every request to a session secured with Kerberos failed with java.lang.ClassNotFoundException: org.springframework.security.crypto.codec.Base64 ^PIVOT-14797.
• A role restricted on a slicing hierarchy holding a single level saw an empty cube instead of the restricted members ^PIVOT-14665. This applies to both atoti.Cube.restrictions and atoti.tables.Tables.restrictions.
• Generating a token on the MCP credentials page no longer fails ^PIVOT-14805. The page is also reachable again directly at /mcp-credentials, without needing /mcp-credentials/index.html.
• username_case_conversion applied to users of other authentication providers ^PYTHON-974. A basic_authentication user registered in lowercase was denied its individual_roles (HTTP 403) when the SSO config used "upper".

​

Added

• atoti-limits ^LIM-2228.
• atoti.Session.chat ^PIVOT-14122.
• atoti.Cube.auto_explain ^PYTHON-824.
• atoti_ai.AiConfig.disclaimer ^PIVOT-14426.
• atoti.SessionConfig.base_path ^PYTHON-917.
• atoti.Cube.dimensions and Dimension ^PIVOT-14220. Use atoti.Dimension.description to provide more context to the Chat.
• atoti.aggregate_provider.AggregateProviders.recommender ^PYTHON-668.
• The history of executed queries is recorded when the observability plugin plugin is enabled and atoti.SessionConfig.user_content_storage is not None ^PYTHON-876. It is accessible in Admin UI (f"{session.url}/admin")‘s Query history tab.

​

Changed

• AiConfig, atoti_ai_amazon_bedrock.ConnectionConfig, atoti_ai_amazon_bedrock.ChatConfig, atoti_ai_openai.ConnectionConfig, and atoti_ai_openai.ChatConfig have been stabilized:
  with tt.experimental({"ai"}):  
      ai_config = AiConfig(
          connection=ConnectionConfig(...),
          chat=ChatConfig(...),
      )
  session_config = tt.SessionConfig(ai=ai_config)
  
• atoti.Table.load(error_handling) defaults to "fail" instead of "log" ^PIVOT-9937. To restore the previous behavior:
  table.load(
      csv_load,
      error_handling="log",  
  )
  
• atoti.NaturalOrder compares digit sequences in strings by their numeric value instead of character by character ^PYTHON-908. String or Object levels default to the new atoti.LexicographicalOrder, which preserves the previous ordering.
• atoti-storage-aws’s commitment policy changed to REQUIRE_ENCRYPT_REQUIRE_DECRYPT ^PIVOT-13185. Existing data on S3 encrypted with older AWS encryption clients (v1 and v2) must be re-encrypted before loading it.
• atoti.Session.start raises a RuntimeError when atoti.SessionConfig.ai is passed but the started server exposes no AI service ^PIVOT-14402. This is most often a license missing the ai-essentials component, which the error message calls out. Such a session previously started successfully with atoti.Session.chat and atoti.Cube.auto_explain silently set to None. To start a session without AI, do not pass ai:
  session_config = tt.SessionConfig(ai=ai_config)  
  session_config = tt.SessionConfig()  
  
  When the ai plugin is enabled but ai is not passed, a RuntimeWarning is emitted instead and the session starts: enabling the plugin alone is not a request for AI.

​

Dependencies

• Upgraded Atoti Server to 6.2.0.
• Upgraded Atoti UI to 5.2.26.
• Bumped minimum required Python version to 3.12 as per NEP 29 ^PYTHON-810.
• Bumped atoti’s minimum required Pandas version to 3.0.3 ^PYTHON-740.
  • DataFrames returned by atoti.Table.query(), atoti.Table.head(), atoti.Session.query_mdx(), and atoti.Cube.query() use datetime64[us] (microsecond precision) instead of datetime64[ns] (nanosecond precision).
  • pyarrow.large_string() columns are supported by atoti.Table.load().
• Bumped atoti-jupyter’s minimum required JupyterLab version to 4.6.0.
• Migrated atoti.Client.http_client from httpx to httpx2 ^PYTHON-903.
• The atoti-parquet plugin is not installed by default anymore. Add its extra to import ParquetLoad:
  uv add atoti
  uv add atoti[parquet]
  

​

Deprecated

• atoti.Hierarchy.dimension_default ^PIVOT-14220. Use atoti.Dimension.default_hierarchy instead:

d, h = cube.dimensions, cube.hierarchies
h["Geo", "Country"].dimension_default = True
d["Geo"].default_hierarchy = h["Geo", "Country"]  

​

Removed

• Support for connecting to Atoti servers older than 6.0.0-M1.
• "Object[]" data type. Use "Object" or the other strongly typed array data types instead.
• atoti.QueryCube.unload_members_from_data_cube() ^PIVOT-14028.
• atoti_ai_openai.ChatConfig.completions_path:
  chat_config = ChatConfig(
      model="gpt-5",
      completions_path="/chat/completions",  
  )
  
• atoti_ai_openai.ChatConfig.http_headers. Use atoti_ai_openai.ConnectionConfig.custom_headers instead:
  chat_config = ChatConfig(
      model="gpt-5",
      http_headers={"X-Header": "value"},  
  )
  connection_config = ConnectionConfig(
      custom_headers={"X-Header": "value"},  
  )
  
• atoti_ai_openai.ChatConfig.retry_max_attempts and the other retry_* attributes. Use atoti_ai_openai.ConnectionConfig.max_retries and atoti_ai_openai.ConnectionConfig.timeout instead:
  chat_config = ChatConfig(
      model="gpt-5",
      retry_max_attempts=3,  
  )
  connection_config = ConnectionConfig(
      max_retries=3,  
      timeout=datetime.timedelta(seconds=30),  
  )
  
• atoti_ai_openai.ConnectionConfig.project_id.
• atoti_directquery_snowflake.ConnectionConfig.array_agg_wrapper_function_name ^PIVOT-10231. Snowflake’s built-in ARRAY_AGG function is faster, so the wrapper UDF is no longer needed.