Dust
Workspace-level platform where users assemble named agents from instructions, knowledge sources, default tools (web search, file creation, image generation, memory), 100+ connectors and remote MCP servers, with explicit agent-as-tool delegation.
Description
Dust is a hosted workspace for building and running AI agents that connect to a company's data, tools and people. Each agent is configured with instructions, a model choice (OpenAI, Anthropic, Google, Mistral), Tools & Knowledge (default tools plus 100+ connector and remote MCP servers) and Data Sources. Dust supports agent-as-tool delegation via the `run_agent` tool, with a documented maximum recursion depth of 4, including a full-handoff mode where the sub-agent takes over the conversation.
Solution
Per-agent runtime: user message arrives, the agent loads its instructions and configured Tools & Knowledge, picks one or more tools per turn (search over Data Sources, default tools, connector tools, MCP-backed tools, or `run_agent` to delegate to another agent), executes, and continues. `run_agent` either calls the sub-agent in a separate conversation and returns the result, or performs a complete handoff where the sub-agent responds directly to the user. Recursion is capped at depth 4.
Primary use cases
- company-wide agent workspace over Notion / Slack / Salesforce / Gmail / etc.
- agent-as-tool delegation to specialised sub-agents
- MCP-extensible tool surface on top of native connectors
- structured-output and Deep Dive agents for long-running work
Open the full interactive page →
Diagram, neighbourhood map, code examples, related patterns and full provenance.