• Dust assistant can now automatically discover and access available tools in the Company Data space • Tools are enabled dynamically on-demand during conversations • Enhanced assistant capabilities for more powerful interactions

• Added 'Sound Studio' tool for generating sound effects
• Added 'Speech Generator' tool to generate discussions, and speech
• Enables podcast generation capabilities

• Deep Dive can now explore and query entire data warehouses (Snowflake, BigQuery) without pre-configuration • Search for tables by name and examine schemas dynamically • Run SQL queries across multiple tables in a single research session • Support for large-scale data exploration (hundreds of tables, thousands of columns)

• Deep Dive sub-agents now receive AI-generated summaries (200-500 tokens) instead of full page content when browsing web pages • Full page content is stored as accessible files for detailed analysis when needed • Enables browsing 20+ pages instead of 2-3 pages within context limits • Improves context efficiency by at least an order of magnitude for research tasks

• @deep-dive can now discover and call MCP tools dynamically during research • Uses toolsets__list to see available tool servers and selects relevant ones (GitHub, Salesforce, Monday, etc.) • Spawns sub-agents with specific tools for focused tasks • Extends capabilities on-demand based on task requirements

• All agents now have improved context management that preserves conversations across more turns • When context window fills up, old tool outputs get pruned while keeping tool calls visible • Maintains reasoning chain so agents remember what they've done • Conversations stay coherent across more consecutive turns • Agents don't lose context even in extended back-and-forth sessions

• Added a new Go Deep Tool that allows agents to hand off complex tasks for comprehensive analysis • Tool is available by default in @Dust agent • Can be added to any custom agent through the agent builder • Enables specialized agents to leverage deep-dive capabilities while maintaining awareness of main agent instructions

• Added a new @deep-dive agent that conducts thorough research across company data, data warehouses, and the web • Uses sub-agents to perform parallel research tasks for extended periods (10+ minutes) • Explores topics iteratively before producing comprehensive answers • Available for complex tasks requiring in-depth analysis and synthesis

• Added option to select 'agent responds in conversation' when using Run Agent MCP tool • Agents can now post their answers directly in the conversation instead of running in background • Available across all workspaces for any user

• Added exploratory search mode with new tools for systematic data navigation • New list tool to browse folders and directory structures • New find tool to search for documents by filename • New cat tool to read complete documents instead of just chunks • New locate_in_tree tool to view document hierarchies • Enhanced search tool with folder-scoped searching capabilities • Toggle available in Agent Builder under 'Advanced settings'