Dust

What it does

Dust is a platform for building AI agents that work with your company's data. It connects to more than 70 tools, including Slack, Notion, Google Drive, GitHub, Salesforce and Zendesk. An agent can read from those sources and answer or act from what it finds.

It supports more than 20 models from providers such as OpenAI, Anthropic, Google and Mistral, so you are not tied to one.

Why you would need it

A general chat assistant knows nothing about your business. You end up pasting context in each time, and the answer is only as good as what you remembered to paste.

You feel the pain when the answer sits in a support ticket, a wiki page and a sales call note, and nobody can find all three. An agent that is connected to those places can.

Where it fits

Dust sits on top of the tools you already use. It reads from them and, depending on set-up, acts in them. The vendor lists use cases by team: code debugging and incident response for engineering, ticket triage and response drafting for support, account research for sales, and brief drafting for marketing.

It sits next to other agent builders such as Lindy and Relevance AI. It replaces ad hoc prompting and shared prompt documents.

What stands out

• A dual-layer permission model that separates what an agent can access from what each user may see.
• SOC 2 Type II certification, US or EU data residency, single sign-on and audit logs.
• Model choice across more than 20 models.
• A free tier to try before committing.

My take

I'd pick it when your team already has knowledge spread across several tools and wants agents that respect permissions. The governance features suggest it was built for companies that cannot let an assistant read everything.

What I'd watch is cost. Usage is credit-based, so a busy agent costs more than a quiet one, and it is harder to predict than a flat seat price. Run a real workload for a week and count credits before you scale it to the team.

Verdict

Pick it when you want agents on your own data with real access control. Skip it when you want the cheapest possible chat tool, or when your data lives in one place already covered by that tool's own AI.