Dust

Dust automatically handles the end-to-end retrieval-augmented generation process, including document ingestion, text chunking, vector embedding generation, and semantic search indexing. This eliminates the need for teams to build and maintain their own vector databases or embedding infrastructure, significantly reducing the engineering effort required to create AI agents grounded in company data.

The platform supports simultaneous access to multiple foundation models from different providers, including GPT-4, Claude, and Mistral. Teams can configure which model each agent uses based on the specific task requirements, balancing factors like response quality, latency, and cost. This prevents vendor lock-in and allows organizations to adopt new models as they become available.

Dust provides a graphical interface for designing multi-step LLM pipelines without writing code. Users can chain together retrieval steps, model calls, and data transformations into structured workflows. This makes it possible for product managers, operations teams, and other non-developers to create and iterate on AI agent behavior while still offering enough depth for technical users.

Native integrations with Slack, Notion, Google Drive, GitHub, Confluence, and other workplace tools allow Dust to continuously ingest and index organizational data. Each connector handles authentication, incremental syncing, and permission mapping, so the data available to AI agents stays current and respects existing access controls from the source systems.

Administrators can define precisely which data sources and tools each AI agent can access, creating role-based boundaries that align with organizational security policies. Combined with audit logging of all agent interactions, this governance layer enables compliance-conscious enterprises to deploy AI assistants while maintaining oversight of how sensitive information is accessed and used.