Dust

Dust starts from a correct observation: the useful version of AI at work is not a general chatbot, it is a set of specialized agents that actually know your company. Founded by early OpenAI and Stripe alumni, Dust lets teams build agents grounded in Notion pages, Slack history, Drive folders, and GitHub repos — then deploy them where people already work.

What we like

The retrieval quality over connected sources is the product's backbone, and it is strong: agents cite the internal documents they drew from, which builds the trust generic chatbots never earn. Being model-agnostic is a real advantage — when a better frontier model ships, your agents upgrade without replatforming. The agent builder is approachable enough that an ops manager, not just an engineer, can ship a working HR-policy answerer in an hour. Permissioning respects source-level access, which is exactly what security teams ask about first.

Where it falls short

At roughly $29 per user per month, company-wide rollouts get expensive quickly, and the value case depends on how good your internal documentation is — agents grounded in a messy Notion are confidently mediocre. Building sophisticated multi-step agents with tool use still lands in technical-user territory despite the friendly UI. And Dust is a scale-up competing with giants; its integration catalog and community are thinner than what surrounds OpenAI's or Anthropic's enterprise offerings.

Verdict

Dust is one of the best choices for 20-500 person companies with well-kept internal knowledge who want many purpose-built agents rather than one chatbot. Run the trial against your three most-asked internal questions; if the answers cite the right docs, roll it out to a pilot team. If your docs are chaos, fix that first — no agent platform can retrieve what was never written down.