Dust
AI assistants for teams - build custom AI agents connected to your company data, powered by the best models
Quick Overview
| Company | Dust |
|---|---|
| Category | AI Chat & Assistants |
| Headquarters | Paris, France |
| EU Presence | Yes - France |
| Open Source | Partially (MIT license) |
| GDPR Compliant | Yes |
| Main Products | Custom AI Agents, Data Connections, Dust Apps, Workspace Platform |
| Pricing | From $29/user/mo |
| Best For | Teams building custom AI workflows |
| Replaces | ChatGPT Teams, Microsoft Copilot |
Detailed Review
Dust has emerged as one of the most compelling European AI platforms for organizations looking to deploy custom AI assistants that are deeply integrated with their existing tools and company knowledge. Founded in 2022 in Paris by Gabriel Hubert and Stanislas Polu, the platform enables teams to build, deploy, and manage AI agents without requiring deep technical expertise. Unlike general-purpose chat interfaces, Dust is purpose-built for the enterprise: it connects securely to a company's data sources, respects granular permission structures, and orchestrates across multiple large language models to deliver contextually rich, accurate responses grounded in an organization's own information.
The company's founding story is rooted in deep Silicon Valley experience brought back to Europe. Stanislas Polu spent seven years in the United States, including time as a software engineer at OpenAI, where he gained firsthand insight into the potential of large language models for workplace productivity. Gabriel Hubert, his co-founder, previously worked at Stripe before returning to France to lead the product team at Alan, the French healthtech unicorn. Together, they recognized that the real value of AI in enterprises would come not from generic chatbots, but from agents that truly understand a company's context -- its documents, conversations, databases, and workflows. This conviction led them to build Dust in Paris, and the platform has since attracted $21.5 million in funding from Sequoia Capital across seed and Series A rounds.
Platform Architecture and Core Concept
At its core, Dust operates as a workspace where organizations create AI agents that are connected to their company's internal data. The platform is model-agnostic, meaning it can orchestrate across leading foundation models including GPT-4, Claude, Gemini, Mistral, and Grok. This flexibility allows organizations to select the best model for each task rather than being locked into a single provider's ecosystem. Agents on Dust can perform multi-step reasoning, processing up to 8 steps (and up to 12 for complex requests) to generate comprehensive answers, making them significantly more capable than simple question-and-answer interfaces.
The architecture revolves around three key layers: data connections, agent configuration, and workspace management. Data connections pull information from the tools teams already use -- Slack, Google Drive, Notion, Confluence, GitHub, Salesforce, HubSpot, BigQuery, Snowflake, and many more. Agent configuration allows admins to define what data each agent can access, which models it should use, and what actions it can perform. Workspace management provides the governance layer, with role-based access controls, SSO/SAML authentication, SCIM provisioning, and audit logs to ensure that sensitive information is handled appropriately.
Data Connections and Integrations
One of Dust's strongest differentiators is the breadth and depth of its data integrations. Workspace admins maintain granular control over exactly which data Dust can ingest from each connected source -- down to specific Slack channels, Google Drive folders, Notion pages, or Confluence spaces. This level of specificity is critical for enterprise environments where different departments or teams may have vastly different data access requirements. The platform supports admin-managed connections for Notion, Google Drive, Confluence, GitHub, Salesforce, HubSpot, BigQuery, Snowflake, Gmail, Google Calendar, Intercom, Zendesk, and more.
Beyond passive data retrieval, Dust agents can take actions within connected platforms. They can interact with Notion to update pages, databases, and add comments. They can work with HubSpot CRM to read, update, and create records. They can query Salesforce data, list objects, and explore object structures. This bidirectional capability transforms Dust from a read-only knowledge assistant into an active participant in business workflows, capable of updating records, creating documents, and triggering downstream processes based on conversational inputs.
Custom AI Agents and Dust Apps
Dust enables organizations to create purpose-built AI agents tailored to specific roles and workflows. A customer support team might build an agent that draws from knowledge base articles, past support tickets, and product documentation to help resolve customer inquiries. A sales team could create an agent that pulls CRM data, meeting transcripts, and competitive intelligence to prepare for prospect calls. An engineering team might configure an agent with access to code repositories, architecture documents, and incident reports to assist with debugging and onboarding. The flexibility to customize agents per department, per use case, or even per project is what sets Dust apart from one-size-fits-all AI tools.
For more advanced users, Dust Apps provide a way to build custom actions and workflows that extend agent capabilities beyond standard retrieval and generation. These can include data transformations, API calls to external services, and complex multi-step automations. The Dust Apps framework allows technical teams to push the platform well beyond its default capabilities, creating bespoke AI-powered processes that would otherwise require custom software development. Dust also supports agent chaining, where one agent can call upon another, enabling sophisticated multi-agent workflows for complex organizational processes.
Security, Compliance, and Data Privacy
Security is a foundational concern for any enterprise AI deployment, and Dust has invested heavily in building trust on this front. The platform holds SOC 2 Type II certification, which it achieved in a remarkably short timeframe, demonstrating robust controls over data security, availability, and confidentiality. Dust is fully GDPR compliant as a French company operating under EU jurisdiction, and it also enables HIPAA compliance for US healthcare organizations that need to process Protected Health Information. Data is encrypted with AES-256 at rest and TLS in transit, meeting the stringent requirements of regulated industries.
Critically, Dust enforces a strict zero-data-retention policy with third-party model providers. When a query is sent to GPT-4, Claude, or any other external model, the data is processed but never stored by the model provider and never used to train their models. This addresses one of the most significant concerns organizations have about enterprise AI adoption. Additionally, Dust offers regional hosting options, allowing organizations to choose between EU and US data residency to meet their specific regulatory requirements. The granular data selection system ensures that admins maintain full control over what data Dust ingests, adding another layer of governance to sensitive information handling.
Workspace Management and Governance
Dust's workspace management layer is designed for the realities of enterprise deployment. Spaces provide a way to organize agents and data by team, project, or access level, ensuring that sensitive information is compartmentalized appropriately. Role-based access controls allow admins to define who can create agents, who can access specific data sources, and who can modify workspace configurations. SSO/SAML integration supports single sign-on through identity providers like Okta, Azure AD, and Google Workspace, while SCIM provisioning automates user lifecycle management as employees join or leave the organization.
Audit logging provides a comprehensive trail of activity across the workspace, which is essential for compliance and security reviews. Enterprise customers also receive priority support and access to dedicated customer success resources. These governance capabilities are what distinguish Dust from simpler AI tools and make it viable for deployment across organizations of hundreds or thousands of employees -- as demonstrated by customers like Doctolib, which successfully rolled out Dust to over 3,000 employees.
Pricing Structure
Dust offers two primary pricing tiers. The Pro plan, designed for small teams and startups, costs 29 euros per user per month (excluding tax). This includes access to advanced AI models, custom agent creation, key data connections, unlimited messages under a fair-use policy, fixed pricing on programmatic API usage, and up to 1 GB per user of data source storage. A 14-day free trial is available for teams to evaluate the platform before committing.
The Enterprise plan is designed for organizations with 100 or more users and offers custom pricing. It includes everything in the Pro plan plus SSO/SAML integration, larger data storage limits, SCIM provisioning for automated user management, regional hosting options (EU or US), priority support, and dedicated customer success management. For organizations that need API access for programmatic usage, Dust also publishes separate API pricing based on the specific models used, allowing developers to integrate Dust's capabilities into custom applications and workflows.
Open Source and Developer Community
Dust maintains an open-source presence through its GitHub organization (dust-tt), where the core platform code is published under the MIT license. This partially open-source approach provides transparency into how the platform works and allows the developer community to inspect, contribute to, and build upon Dust's codebase. The company also maintains an official JavaScript/TypeScript SDK for building integrations, and contributes to the broader open-source ecosystem, including work on Model Context Protocol (MCP) servers and the Firecrawl web scraping tool.
The decision to open-source portions of the platform under the permissive MIT license reflects the founders' roots in the open-source community and aligns with broader European values around transparency and digital sovereignty. While the hosted SaaS product is where most customers interact with Dust, the open-source availability means that technically sophisticated organizations can audit the code, understand exactly how their data is being handled, and even contribute improvements back to the platform.
Customer Traction and Real-World Adoption
Dust has gained significant traction since its launch, reaching over 2,000 organizations and generating $7.3 million in annual recurring revenue by mid-2025. Notable customers include Doctolib (Europe's largest health tech platform, with 3,000+ Dust users), Alan (French healthtech unicorn), Wakam (insurance company operating through 100 distributor partners across 32 countries), Blueground (which used Dust to boost customer satisfaction and reduce support resolution time), and Clay (which leveraged Dust to scale its go-to-market team). This diverse customer base spanning healthcare, insurance, real estate, and SaaS demonstrates the platform's versatility across industries.
The company itself has grown efficiently, reaching its revenue milestones with a team of approximately 66 employees -- a testament to the lean operational approach favored by the founders. Dust passed the $1 million and $2 million ARR milestones during 2024 before accelerating rapidly through 2025, with ambitions to quintuple its growth. This capital-efficient growth, backed by Sequoia Capital's expertise in scaling enterprise software companies, positions Dust well for continued expansion across European and global markets.
Competitive Position
In the enterprise AI agent space, Dust competes with several categories of products. Against general-purpose AI assistants like ChatGPT Teams and Microsoft Copilot, Dust differentiates through its deeper data integration capabilities, model-agnostic architecture, and granular permission controls. Where ChatGPT Teams offers a conversation interface with limited data connectivity, and Microsoft Copilot is tightly coupled to the Microsoft 365 ecosystem, Dust works across the full spectrum of enterprise tools regardless of vendor. Its European heritage and GDPR-native compliance also give it a meaningful advantage for organizations that prioritize data sovereignty.
Against AI workflow automation platforms like n8n and other no-code tools, Dust offers a more conversational and agent-centric approach that is easier for non-technical users to interact with. Against enterprise search and knowledge management tools, Dust provides the added value of generative AI that can synthesize, analyze, and act on information rather than merely retrieving it. The combination of conversational AI, deep data integration, multi-model orchestration, and enterprise governance in a single platform is relatively rare, giving Dust a distinctive position in the market.
Limitations and Considerations
Despite its strengths, Dust has certain limitations that prospective users should consider. The per-seat pricing model of 29 euros per user per month can become expensive when deploying across large organizations, potentially pushing teams toward the custom-priced Enterprise tier sooner than expected. The 1 GB per user data source storage limit on the Pro plan may be restrictive for organizations with large document repositories or extensive data needs. Some users have reported that the platform can be challenging for novices to configure initially, given the breadth of features and customization options available.
The Dust browser extension currently has a dependency on Chrome, which excludes users who prefer Firefox, Safari, or other browsers. Working with very large or complex multi-source datasets can sometimes present challenges in terms of retrieval accuracy and response quality. As a Series A startup with 66 employees, Dust's support resources and documentation, while growing, may not yet match those of larger, more established enterprise software providers. These are, however, typical growing pains for a rapidly scaling startup and are likely to be addressed as the company continues to mature and expand its team.
Future Outlook and European Significance
Dust represents a significant entry in the European AI landscape -- a French company building enterprise-grade AI infrastructure that competes directly with products from American tech giants. As organizations increasingly seek AI platforms that respect European data protection standards and support digital sovereignty, Dust is well-positioned to capture demand from companies that want the productivity benefits of AI agents without the regulatory and privacy risks of US-based platforms. With strong backing from Sequoia Capital, a growing roster of marquee European customers, and an open-source ethos that aligns with European values, Dust is one of the most promising AI startups to emerge from Paris in recent years.
Alternatives to Dust
Looking for other European AI Chat & Assistants solutions? Here are some alternatives worth considering:
Frequently Asked Questions
Dust is a French AI agent platform that enables teams to build custom AI assistants connected to their company's data. It connects to tools like Slack, Google Drive, Notion, Confluence, Salesforce, and more, allowing AI agents to retrieve and act on company knowledge. The platform is model-agnostic, orchestrating across GPT-4, Claude, Mistral, Gemini, and Grok to deliver contextually grounded responses. Agents can perform multi-step reasoning (up to 12 steps for complex requests) and take actions within connected platforms.
Unlike ChatGPT Teams, which offers limited data connectivity, and Microsoft Copilot, which is tied to the Microsoft 365 ecosystem, Dust works across the full spectrum of enterprise tools regardless of vendor. Dust is also model-agnostic, letting you choose between GPT-4, Claude, Mistral, and others rather than being locked into one provider. Additionally, Dust is GDPR-compliant by design as a French company, offers EU data residency, and provides granular permission controls that give admins precise control over what data each agent can access.
Yes, Dust is fully GDPR compliant as a French company operating under EU law. It also holds SOC 2 Type II certification and enables HIPAA compliance. Data is encrypted with AES-256 at rest and TLS in transit. Critically, Dust enforces a zero-data-retention policy with third-party model providers -- your data is never stored by or used to train external models. Enterprise customers can choose between EU and US data residency for regional hosting.
Dust supports admin-managed connections to a wide range of enterprise tools including Slack, Google Drive, Gmail, Google Calendar, Notion, Confluence, GitHub, Salesforce, HubSpot, BigQuery, Snowflake, Intercom, Zendesk, and more. Admins have granular control over exactly which data Dust can ingest -- down to specific Slack channels, Drive folders, and Notion pages. Agents can also take actions in connected platforms, such as updating Notion pages or creating HubSpot records.
Dust offers two pricing tiers. The Pro plan costs 29 euros per user per month (excluding tax) and includes advanced AI models, custom agent creation, key data connections, unlimited messages under fair-use policy, and up to 1 GB per user of data storage. A 14-day free trial is available. The Enterprise plan offers custom pricing for organizations with 100+ users and adds SSO/SAML, SCIM provisioning, larger storage, regional hosting, and priority support. API pricing for programmatic usage is also available separately.
Dust was founded in 2022 in Paris, France, by Gabriel Hubert and Stanislas Polu. Stanislas previously spent seven years in the US working as a software engineer at OpenAI, while Gabriel worked at Stripe before leading the product team at Alan, the French healthtech unicorn. The company has raised $21.5 million in funding from Sequoia Capital across seed and Series A rounds and operates with a team of approximately 66 employees.
Dust is partially open source. The core platform code is published on GitHub (github.com/dust-tt/dust) under the MIT license, allowing developers to inspect, audit, and contribute to the codebase. The company also maintains an official JavaScript/TypeScript SDK and contributes to open-source projects like Model Context Protocol servers and Firecrawl. Most customers use the hosted SaaS product, but the open-source availability provides transparency and enables technical organizations to audit how their data is handled.
Dust is model-agnostic and supports orchestration across multiple leading foundation models. Currently available models include GPT-4 from OpenAI, Claude from Anthropic, Gemini from Google, Mistral models, and Grok 3 / Grok 3 Mini from xAI. This flexibility allows organizations to select the best model for each use case, switch between models as capabilities evolve, and avoid vendor lock-in with any single AI provider.
Dust is particularly well-suited for teams that need AI assistants grounded in company-specific knowledge. Common use cases include customer support (agents drawing from knowledge bases and past tickets), sales enablement (CRM data and competitive intelligence), engineering (code repositories, architecture docs, incident reports), HR (policy documents and onboarding materials), and operations (meeting transcripts, project documentation). Companies like Doctolib, Alan, Wakam, and Clay use Dust across thousands of employees for diverse workflows.
Key limitations include per-seat pricing that can add up for large teams, a 1 GB per user data storage cap on the Pro plan, and initial complexity for non-technical users given the breadth of configuration options. The browser extension currently only works on Chrome. Working with very large, multi-source datasets can sometimes challenge retrieval accuracy. As a Series A startup, Dust's support resources are growing but may not yet match those of larger enterprise software vendors. These are evolving limitations that the team is actively addressing.