Building
Start from a conversation, a template, or scratch. Write instructions with help from an assistant called Sidekick, pick tools and knowledge, test in a preview, and publish to the workspace.
Compare
Dust
a platform for building, sharing, and running AI agents connected to company knowledge, tools, and teams
Who they are and what they are good at
Dust, at dust.tt, describes itself as multiplayer AI for human-agent collaboration. Teams build agents in a shared workspace, and people use them in the web app and a browser extension.
Knowledge and tools
Connectors to Slack, Notion, Google Drive, GitHub, Salesforce, Zendesk, and more, plus remote MCP servers. Admins choose what data syncs, and access follows spaces.
Models
More than 20 models, including GPT, Claude, Gemini, and Mistral.
Triggers
Schedules written in plain language, and webhook triggers from tools such as GitHub, Jira, and Zendesk.
Pricing is per seat with monthly credits, from a free tier to Pro and Max, plus custom Enterprise plans. Dust is SOC 2 Type II certified and offers US or EU data residency. For research, support answers, sales prep, and writing across SaaS knowledge, it is a very good fit.
Where Dust stops for messy back-office work
Dust reaches software through connectors, MCP, and the page open in your browser. Back-office procedures often live somewhere else.
Desktop apps
Dust does not document a way to operate installed desktop software, such as a Windows ERP or an app delivered through Citrix.
Browser actions are attended
The extension can click and type in the page you have open, after a permission prompt. That suits a person at the browser more than an overnight run.
Exact math
Tying a remittance to open invoices to the cent should be a rule, not a model's judgment.
Evidence is a conversation
Scheduled runs appear in the creator's history unless they are created in a shared Pod. Audit logs are an Enterprise feature.
LaunchAI operates any app through the screen, on the user's machine with their existing logins, and records every action. See Product →.
Side by side
| Dust | LaunchAI | |
|---|---|---|
| Built for | Team agents on company knowledge and SaaS tools | Recurring back-office procedures on any screen |
| Who builds | Workspace members, with help from Sidekick | The process owner, by talking and showing |
| How it reaches software | Connectors, MCP, and the open browser tab | The screen, through the Vision Layer; no APIs needed |
| Where it runs | Hosted by Dust in the US or EU; single-tenant on Enterprise | The user's machine or an enrolled agent PC |
| Desktop apps with no API | Not documented | Any app on the machine |
| Schedules | Plain-language schedules and webhooks | Schedules, triggers, and a run console |
| Exact math | Model judgment | Deterministic rules, no model call |
| Evidence | Conversation history; audit logs on Enterprise | Before-and-after screenshots of every action |
| Model choice | 20+ models | OpenAI, Anthropic, Azure, Bedrock, or self-hosted, per step |
| Cost unit | Seats with monthly credits | License, plus your model account |
When to choose Dust
- A support team wants an agent that answers from Zendesk, Notion, and Slack history, and drafts the reply.
- Sales and operations want shared agents for research and writing, with access set by space and a choice of models.
When to choose LaunchAI
- Cash application: download the lockbox file from a bank portal, match remittances to open invoices in a desktop ERP, post exact amounts, and stop for review on anything short-paid.
- The job runs on a schedule, on screens with no connector, under the AR specialist's own login.
- An auditor asks how a payment was applied, and the answer should be a screenshot.
Better together
Use Dust where the knowledge lives: policy answers, account research, and drafts. Use LaunchAI where the clicks live: the ERP, the bank portal, the PDF. LaunchAI's run history exports in an audit-friendly format. Save it to a drive Dust is connected to, and a Dust agent can answer questions about last month's runs.
FAQ
Can Dust agents click through websites?
Through the browser extension, yes, in the page you have open and after a permission prompt. Dust turns page interaction off where a dedicated tool exists, such as Notion or Gmail.
Can Dust run an agent on a schedule?
Yes. Schedules are set in plain language in the Agent Builder. By default, each run appears only in the creator's history unless it is created in a shared Pod.
Does Dust work with desktop software like a Windows ERP?
Not that Dust documents. It connects through connectors and MCP, so a system with neither needs another route.
Do we have to choose?
No. They cover different work: Dust for knowledge and writing, LaunchAI for screen procedures that repeat.
Can LaunchAI use our existing model accounts?
Yes. Model steps run on the provider you configure, and deterministic steps make no model call.
No prompts. No babysitting. No Integrations.