How to Use Dust to Build Custom AI Assistants for Your Team
Dust is a platform for building AI assistants connected to your team’s data and apps. You create an assistant, attach sources like Notion, Google Drive, or a database, pick a model, and share it to Slack or a web link. The entry plan is paid per seat, with a free trial; pricing scales by workspace size and connector use.
The appeal is that a non-engineer can assemble a working internal assistant without writing a backend. The risk is locking a half-built assistant in place and calling it done.
What is Dust?
Dust is an assistant builder and runtime. Assistants are prompt plus tools plus knowledge sources, packaged so everyone in a workspace can use the same version.
It ships connectors so an assistant can read specific documents rather than relying on a model’s memory. That grounding is what makes answers useful instead of generic.
Is Dust free?
There is a free trial, then paid plans billed per seat. Cost depends on workspace size and how heavily assistants use connectors and models.
Compare it to building on raw APIs: Dust removes engineering time but adds a per-seat cost. For a team of five, the tradeoff usually favors Dust; for a solo developer with one use case, an API call is cheaper.
How do you build your first assistant?
- Create a workspace and invite teammates who will use the assistant.
- Click New assistant and give it a name and a system prompt describing its job.
- Add knowledge: connect Notion, Drive, or upload files, and scope it to specific folders.
- Pick a model — a fast one for drafting, a stronger one for analysis.
- Test in the preview pane, then share to Slack or copy the web link.
Keep the first assistant narrow. “Answer questions about our onboarding docs” beats “help with everything,” and it is far easier to tell whether it works.
How should you scope knowledge sources?
Point Dust at the smallest set of documents that answers the question. Connecting an entire company Drive gives the assistant more to search and more irrelevant matches to trip over.
If an assistant handles HR questions, feed it the handbook folder, not the whole drive. Retrieval quality drops when the corpus is wide and noisy.
What can assistants connect to?
| Type | Example | Use case |
|---|---|---|
| Docs | Notion, Google Drive | Internal Q&A |
| Code | GitHub | Explaining repos, PR questions |
| Chat | Slack | Fielding questions in channel |
| Web | Browsing tool | Current info when needed |
Where do teams get this wrong?
Two places. First, permissions — an assistant reading everything can surface a sensitive doc to someone who should not see it, so verify source-level access before sharing broadly. Second, no owner. An assistant nobody maintains drifts as the underlying docs change.
Assign one person per assistant to review it monthly. And be honest about output quality: retrieval assistants confidently cite the wrong passage when two documents conflict, so keep a human in the loop for anything decision-critical.
Who should use Dust?
Teams with a body of internal knowledge and no engineer to spare — support, ops, and people teams fielding repeated questions. Skip it if you have one question and one document; a bookmark is faster.
Quick Summary
- Dust builds shared AI assistants from a prompt, connected data sources, and a model.
- Pricing is per seat after a free trial; workspaces scale by size and connector usage.
- Scope knowledge sources narrowly — feeding a whole drive lowers answer quality.
- Verify source permissions before sharing, and assign an owner to review each assistant.
