Dust Review

Most AI assistants are blind to the thing that makes them useful at work: your company's own knowledge. Dust's pitch is to fix that, building agents that read across Slack, Drive, Notion and your other systems, on your terms.

It lets teams build and run AI agents with company context through 70+ connectors and 20+ models, with a permission model, SCIM, audit logs and analytics. It sits among agent platforms like Relevance AI, at the enterprise-knowledge end.

This review covers how Dust runs company agents, which teams it suits, how it's priced, and the data-governance duties that come with a tool that spans your systems.

Dust, The Internal-Agent Brief

• Dust builds agents: AI agents grounded in your company's knowledge.
• It connects widely: 70+ connectors and 20+ frontier and open models.
• It's built for control: dual-layer permissions, SCIM, audit logs, SOC 2.
• Mind the scope: it operates across your systems, so governance is central.

Dust Internal-Agent Facts

Building AI Agents On Your Company's Knowledge

Dust's core idea is to make AI useful at work by grounding it in company knowledge. Instead of a blank chatbot, you build agents that pull relevant context across Slack, CRM, docs and other systems through 70+ connectors, so answers reflect your actual processes and data rather than the open internet.

It's model-flexible, 20+ frontier and open models from OpenAI, Anthropic, Google and Mistral that teams can choose between, and ships the surrounding apparatus: a Chrome extension, Frames, a developer platform and an API to build agents into workflows.

Crucially for its buyer, it leads with governance: a dual-layer permission model, SCIM provisioning, audit logs and analytics, SOC 2 Type II, US/EU data residency and custom retention. Its framing, 'use AI on your terms without giving up control of your data' and 'granular enough for your CISO, invisible to everyone else', is pitched at exactly the security review an enterprise rollout triggers.

The honest framing is that Dust is a serious, well-designed enterprise agent platform, and grounding agents in company knowledge with real permission controls is the right architecture for business use. The duties follow from the power: an agent that reads across your systems is only as safe as its permissions, so the dual-layer model and audit logs must be configured and reviewed, not assumed; data residency and retention should be confirmed against your obligations; and the '3,000+ organizations' figure is a vendor claim. Used with governance owned by someone, it's a capable platform.

Dust builds AI agents grounded in your company's knowledge with CISO-grade controls; the architecture is right for enterprise, provided someone owns the permissions and data governance.

How Does Dust Run Company Agents?

Dust pairs company-context agents with model choice and enterprise governance.

Connectors, Models And Context

Agents read across 70+ connected systems and run on a choice of 20+ frontier and open models.

A platform like Stack AI builds AI workflows and apps; Dust's angle is multiplayer agents grounded in company knowledge, a context-first job.

Permissions, Governance And Your Data

A dual-layer permission model, SCIM, audit logs, API and a Chrome extension wrap it for enterprise.

An automation tool like Activepieces wires apps together with flows; Dust centres on knowledge-grounded agents, so match it to whether you need agents or pipelines, and own the governance either way.

Which Internal Teams Dust Suits

Dust suits enterprise departments, sales, support, marketing, engineering, data, legal, HR and IT, that want AI agents grounded in their own tools and documents rather than a generic assistant.

It suits organisations that need the governance apparatus, permissions, SCIM, audit logs, data residency, to pass a security review, and that have someone to own agent configuration.

It suits less well small teams or individuals who'd find an enterprise platform and its sales-led rollout heavy, and organisations without the connectors or the appetite to govern an AI that reads across their systems.

A fair way to place it: if you're an organisation that wants capable, context-grounded agents and can own the permission and data governance, Dust is a strong, well-architected choice; if you're a small team wanting a quick assistant, or no one will own the governance, weigh it carefully, because the platform's strength, reach across your systems, is also the thing that must be controlled. Assign a clear owner for connectors and permissions, start with a bounded, high-value use case, and review the audit logs before you widen access.

How Dust Charges Enterprises

Dust is free to try and then sold to enterprises, sales-led ('try for free', 'contact sales', 'request a demo'), with shared workspace credits and volume pricing mentioned but no amounts on the captured page. We do not publish prices, which vary; confirm the plan, credits and data terms with Dust.

The value question is productivity across departments versus platform cost and governance effort: for an organisation that deploys agents widely and grounds them in its knowledge, the payoff can be large, but it's an investment that needs ownership.

The sensible approach is to trial Dust on a bounded, high-value use case, connect a limited set of systems, verify the permissions and audit trail behave, confirm data residency and retention against your obligations, and expand only once the governance holds.

Standing Up One Agent In Dust

Because Dust reads across your systems, trial it on a bounded use case with governance owned.

1. Trial it on one bounded, high-value use case first.
2. Connect a limited set of systems and build a focused agent.
3. Assign an owner for connectors and the permission model.
4. Verify the dual-layer permissions and audit logs behave as expected.
5. Confirm data residency and retention against your obligations.
6. Pass/fail: Dust is worth it if grounded agents help and governance holds.
7. Treat the organisation-count figure as vendor-stated.

Dust Internal-Agent Terms

AI agent

A configurable AI that can use tools and company context to carry out tasks, not just answer questions.

Connector

An integration that lets Dust read data from a system like Slack, Drive or Salesforce, subject to permissions.

SCIM

A standard for automatically provisioning and de-provisioning user accounts and access from your identity system.

Data residency

Where your data is stored and processed (e.g. US or EU), which can matter for compliance obligations.