Skip to main content
Dust is a platform powered by the top frontier models on the market, from providers including OpenAI, Anthropic, Google, and Mistral, alongside open-source models. It’s designed to help teams work better with AI: agents help anyone in the company answer questions in your company’s context, draft documents, analyze data, and act in connected systems. Our mission: we think people don’t get up to go to work to use the enterprise search portal or to have a better knowledge solution but to get stuff done. We build Dust to make work work better. You can read more here. The developer platform lets you build on Dust programmatically through the Dust API, the JavaScript SDK, and the Dust CLI.

​
Agents

An agent combines a model with instructions, tools, and company knowledge. Every workspace includes the default @dust agent, and members can build custom agents for specific tasks, like answering support questions or generating SQL from company data. Several agents can work together in the same conversation.

​
Models

Every agent runs on a model. The model picker offers specific models from each provider plus three auto models (Basic, Standard, Premium) that Dust benchmarks and keeps up to date. Admins can restrict which models members can use with model access tiers.

​
Conversations

You interact with agents through conversations, where you ask questions or make requests. Share a conversation to let colleagues join and work with you and the agents. Create a new conversation for each new topic to keep agents focused.

​
Workspaces

A workspace is where members talk to agents and customize Dust for their team’s needs. Admins invite members, assign roles (Admin, Manager, Member), and organize data access through open or restricted spaces.

​
Data sources

Data sources make company knowledge available to agents:
  • Connections: managed integrations (Google Drive, Notion, Slack, Confluence, and more), synced automatically.
  • Tools: access and act on third-party systems (Slack, Salesforce, Gmail, HubSpot, and more) via MCP servers.
  • Public websites: content ingested with the Dust web crawler.
  • Folders: static documents and tables, uploaded manually or by API.
  • Custom connections: connections you build yourself with the Dust API or ready-made scripts.
Learn more in the data sources overview.

​
Synchronizing

Connections sync automatically, and updates usually appear in Dust within a few minutes. Admins have granular control over what is synced, down to specific Slack channels, Google Drive folders, or Notion pages.

​
Retrieve

Agents retrieve information from data sources using Knowledge tools:
  • Search: semantic search across selected data sources (RAG).
  • Include data: include documents exhaustively, most recent first, up to the context window limit.
  • Query tables: turn a question into a SQL query against structured data.
  • Extract data: pull structured information following a schema you define.