Review

Dust: Enterprise agents for the company sprawl

Dust is a strong choice for enterprises that want internal AI agents tied to company systems, but its pricing, governance overhead, and operational bias make it a poor fit for casual users.

Last updated April 2026 · Pricing and features verified against official documentation

Most enterprise AI products still behave like chat windows wrapped in procurement language. Dust is more honest than that. It is trying to be the operating layer for internal work, where assistants are grounded in company data, tied to specific permissions, and deployed across Slack, the web, Teams, Chrome, Sheets, and APIs. Founded in 2023 by Gabriel Hubert and Stanislas Polu, and based in Paris under Permutation Labs, the company has moved fast enough to turn a startup story into a real enterprise product.

That speed matters because Dust is not a toy assistant for curious employees. TechCrunch reported early enterprise customers using it heavily, and VentureBeat later wrote that the company had reached $6 million in annual revenue by 2025. That combination tells you where the product fits: Dust is built for companies that want AI embedded in the work itself, not parked in a sidebar.

The best case for Dust is that it understands what enterprise AI actually needs: context, permissions, and control. If your organization lives inside Slack, Notion, Google Drive, GitHub, Zendesk, Salesforce, and a dozen other systems, a generic chatbot is mostly a liability. Dust can be a serious answer when the job is to turn that sprawl into usable internal agents without asking employees to adopt a new way of working.

The case against it is just as clear. Dust is not cheap, not lightweight, and not especially forgiving of weak data hygiene. The platform works best when admins can curate sources, teams can think clearly about governance, and buyers are willing to pay for human usage and machine usage separately. In short, Dust is a strong enterprise product for teams that already know they need one. Everyone else should look somewhere simpler.

What the Product Actually Is Now

Dust is a workspace for building specialized agents on company knowledge and tools. The current product is not just a chat surface. It includes agents, triggers, persistent memory, and programmatic usage via API, Google Sheets, and Zapier, while deployment surfaces span the web app, Slack, browser extension, Microsoft Teams, and Raycast. That matters because the product is designed to keep employee use and automation use in the same control plane.

The company is also explicit about model choice. The pricing and security pages list GPT-5, Claude, Gemini, and Mistral, and the security page says customers can choose trusted providers with no data fed into training. Dust’s own framing is revealing: it wants to be the AI operating system for the enterprise, which sounds ambitious because it is.

Strengths

Specialized assistants are a better fit than one universal bot. Dust does not force every request through the same generic interface. It is built around custom agents that can be scoped to different teams, data sources, and actions, which is the right shape for support, HR, sales, engineering, and operations. TechCrunch’s coverage of early customers suggested the same pattern, with teams like Qonto and Alan building multiple assistants instead of pretending one bot could cover everything.

It lives where work already happens. Dust is not asking employees to abandon Slack, Teams, or the browser and go visit a separate AI destination. That matters more than it sounds, because most internal tools die when they add one more place to check. By showing up inside the collaboration surfaces teams already use, Dust reduces the friction between asking a question and doing something with the answer.

Human use and programmatic use are not mashed together. The product separates interactive employee usage from API- and automation-driven usage, which is the right model for a platform that expects both people and systems to call the same agents. That separation makes the product easier to govern than a loose collection of scripts, and it makes the commercial model more honest than fake flat-rate automation pricing.

The security story is specific, not decorative. Dust says customer and personal data are not used to train foundation models, the security page lists SOC 2 Type II, GDPR, and HIPAA support, and the platform can host in either the EU or the US. That is not a guarantee that every deployment will satisfy every legal or procurement team, but it is a serious baseline for an enterprise product.

Weaknesses

It depends heavily on company hygiene. Dust can only be as useful as the data and permissions behind it. If internal documentation is stale, if access controls are messy, or if a team has not decided which sources should be authoritative, the assistants inherit that confusion. A 2026 Cybernews review based on hands-on testing found the product flexible and easy to use but still pointed to friction with large, multi-source data sets and governance.

The bill becomes more complicated once work is real. The public pricing is simple only at the surface. Human use has one price, programmatic usage is metered separately, and the enterprise tier adds custom pay-per-use billing on top of everything else. That is normal for enterprise infrastructure, but it means Dust is not the kind of tool you buy casually and forget about.

It is built for organizations, not individuals. Even the self-serve plan is framed around small teams, and the enterprise tier starts at 100 members. That is a clue about the product’s center of gravity. Dust is not trying to be a personal productivity app that quietly grows into a company standard. It is a company standard from the start, which means setup, governance, and change management are part of the purchase.

Pricing

Dust’s pricing tells you immediately who the product is for. Pro is the only self-serve tier, and at €29 per user per month it is reasonable for a small team that wants to test the system properly. The 14-day trial lowers the barrier to entry, but the 1GB-per-user source cap and the separate charges for programmatic usage mean the real cost starts to matter once the product becomes operational.

Enterprise is the actual platform tier. It adds multi-workspace management, SSO, SCIM, dedicated support, regional hosting, and custom programmatic billing, but it is custom-priced and aimed at 100+ users. That makes the commercial story straightforward: Pro is for pilots and small deployments, while Enterprise is for organizations that expect Dust to become part of how work runs.

The practical takeaway is that Dust is priced like infrastructure with a seat-shaped front door. If you want a lightweight monthly assistant, the pricing will feel busy. If you want a governed internal AI layer that can be used by people and systems, the model is coherent.

Privacy

Dust’s privacy posture is one of the stronger arguments for the product. The company says foundational model providers are prohibited from using customer or personal data for training, the security page says it uses zero data retention with trusted providers, and the same page says data can be hosted in the EU or the US. That is the sort of language enterprise buyers expect to see before they allow an AI system anywhere near internal content.

The platform privacy policy is also fairly explicit about how Dust handles its own user data. Permutation Labs, the Paris-based company behind Dust, describes itself as the data controller for its own platform activity while acting as a processor for customer content. It also says user data can be used for marketing, lead qualification, and internal development, and that transfers outside the EEA can happen under standard contractual clauses. That is not unusual for SaaS, but it is worth saying plainly: Dust has strong model-training boundaries, not magical data invisibility.

Who It’s Best For

• The enterprise operations team with lots of internal tools and repeated questions. It needs assistants that answer from current docs and systems, and Dust wins by keeping permissions and sources close to the agent itself.
• The support, HR, or sales team that wants multiple specialized assistants instead of one general bot. It gets a better fit from Dust because each assistant can be scoped to the right data and deployed where work already happens.
• The platform or security team that cares about regional hosting, SSO, and data-training boundaries. Dust is one of the clearer options when governance matters as much as output quality.

Who Should Look Elsewhere

• Teams whose main problem is cross-company search should start with Glean. Dust is broader, but Glean is more directly built around retrieval across enterprise sprawl.
• Buyers who want workflow orchestration and agent governance more than company knowledge should compare Relevance AI. Dust is stronger on internal context; Relevance AI is more obviously structured as an operating layer for agent work.
• Microsoft-centric organizations should check Copilot Studio first. Dust is cross-platform, but Copilot Studio fits better if the real requirement is governance inside Microsoft 365 and Power Platform.

Bottom Line

Dust is one of the more convincing arguments for treating internal AI as infrastructure rather than as a novelty assistant. It is strongest when a company wants a governed layer on top of its own systems and a place to run both human and automated use cases.

That also defines the boundary. Dust is for organizations that can tolerate operational complexity in exchange for control and repeatability. If that describes your environment, it belongs on the shortlist. If you wanted something that behaves like a consumer app with a business plan, it does not.