Manage data products through a dedicated API surface that mirrors the existing Workspaces API. Data products provide the same capabilities — data assets, semantic models, verified queries, system prompts, front page configs, chat, skills, and compositions — under a streamlined data-products URL namespace. This enables API consumers to work with data products directly without needing to understand the internal workspace abstraction.
Token counts and cost fields have been removed from the Chat Threads API and replaced with BCU (Bobsled Compute Units) — a normalized compute metric that provides a consistent measure of AI resource usage regardless of the underlying model.
Filter chat thread search results by date range and browse recent threads without a text query. Date filters use updatedAt so that resumed conversations appear in results for the time period they were last active, not just when they were first created.
Search chat threads with an enhanced agent mode that returns larger content snippets, filters out tool metadata, and includes turn numbers for precise conversation navigation. A new thread context endpoint lets you fetch full conversation content around a specific turn.
Workspace creation and data asset update operations now support significantly larger payloads. Previously, both endpoints capped arrays at 100 items, causing opaque "Validation failed" errors for customers with large data catalogs. The per-request limit has been raised to 500, and the system automatically batches operations internally so workspaces with thousands of tables work seamlessly.
Monitor data source health and update configurations through new API capabilities. The list endpoint now includes real-time connection status, and a new PATCH endpoint allows updating data source names and credentials without recreating the resource.
Connect to Google BigQuery as a data source alongside Snowflake, PostgreSQL, and Databricks. Organizations using BigQuery can now leverage Bobsled AI's natural language to SQL capabilities against their BigQuery projects using GCP Workload Identity Federation for secure, keyless authentication.
Connect to Databricks SQL Warehouses as a data source alongside Snowflake and PostgreSQL. This enables organizations using Databricks to leverage Bobsled AI's natural language to SQL capabilities directly against their lakehouse data.
The run-sql endpoint now supports a codeType parameter to distinguish between SQL and Python code execution. This enables M2M clients to execute Python code while maintaining SQL validation for regular queries.
The account API response now includes a reference to the account's observability workspace, enabling clients to detect when observability features are available.
