Dust

What Is Dust?

Picture this: a product manager at a 60-person SaaS company needs to draft a customer response about a feature request. The context is spread across a Notion document describing the product roadmap, a Slack thread where engineering discussed feasibility, a GitHub issue tracking the implementation, and a Google Drive folder with customer research from last quarter. Assembling this context manually takes 30 minutes of tab-switching, searching, and copying. The response takes 10 minutes to write.

Dust exists to collapse that 30 minutes into 30 seconds.

Founded in Paris in 2022 by former engineers from Stripe and Alan (the French health insurance startup), Dust is an AI assistant platform that connects to your company's internal tools and creates AI agents grounded in your actual organisational knowledge. It is not a generic chatbot. It is a retrieval-augmented generation (RAG) system that indexes your Notion pages, Slack messages, GitHub repositories, Google Drive documents, Confluence wikis, Intercom conversations, and more -- then uses that indexed knowledge to answer questions, draft content, and assist workflows with context that a standalone LLM could never possess.

The platform supports multiple LLM backends -- GPT-4, Claude, and Mistral -- letting teams choose the most appropriate model for each use case without vendor lock-in. As a French SAS based in Paris with EU data processing, Dust provides a sovereign alternative to US-only AI assistant platforms like Microsoft Copilot or ChatGPT Enterprise.

Dust is designed for teams, not individuals. The Pro plan starts at EUR 29 per user per month for up to 50 users, with Business and Enterprise tiers for larger organisations. This pricing positions it as a productivity investment -- justified if each team member saves meaningful hours per week by having an AI assistant that can actually access and synthesise company knowledge.

Key Features

Custom AI Agent Builder

Dust's most powerful feature is the ability to build custom AI agents tailored to specific workflows. An engineering team can create an agent that answers questions about the codebase by referencing GitHub repositories and technical documentation. A support team can build an agent that drafts customer responses by referencing the knowledge base, past tickets, and product documentation. A sales team can create an agent that generates prospect briefs by pulling from the CRM and company research.

The agent builder uses a no-code configuration interface. You select data sources, define the agent's role and instructions, choose the LLM backend, and deploy it to your team workspace. Agents can use tools -- performing web searches, running calculations, or calling APIs -- extending their capabilities beyond simple question-answering. The ability to create multiple specialised agents rather than one general-purpose chatbot is what distinguishes Dust from simpler AI assistant products.

15+ Native Data Source Connectors

Dust connects natively to over 15 enterprise tools: Notion, Slack, GitHub, Google Drive, Confluence, Intercom, Microsoft Teams, Linear, and others. These connectors index your data continuously, keeping the AI's knowledge base current as documents are updated, messages are posted, and code is committed.

The quality of Dust's responses depends directly on the quality of the connected data. Well-structured Notion pages, detailed Slack threads, and thorough GitHub documentation produce excellent AI responses. Poorly organised data -- scattered across undocumented Slack channels, outdated wiki pages, and abandoned documents -- produces mediocre results regardless of the LLM's capabilities. Dust surfaces this data quality issue rather than creating it, but organisations should expect to invest in information hygiene alongside Dust adoption.

Multi-LLM Support Without Vendor Lock-In

Dust supports multiple LLM providers including OpenAI (GPT-4), Anthropic (Claude), and Mistral, allowing teams to select the best model for each task. A customer support agent might use Claude for its longer context window and nuanced tone. A code analysis agent might use GPT-4 for its stronger technical capabilities. A French-language agent might use Mistral for its native European language understanding.

This multi-model approach provides meaningful flexibility and eliminates single-vendor dependency. If OpenAI changes pricing, degrades quality, or faces regulatory issues, your agents can be switched to alternative models without rebuilding them. For European companies concerned about US AI provider dependency, the ability to route through Mistral -- a French company -- adds an additional sovereignty option.

Permission-Mirrored Data Access

One of Dust's most important features for enterprise adoption is permission-mirrored data access. When an employee queries a Dust agent, the system only returns information from data sources that employee is authorised to access in the original tools. If an engineer does not have access to a private HR Slack channel, the AI agent will not surface information from that channel -- even if the data is indexed.

This permission mirroring is essential for organisations where different teams have different data access levels. Without it, AI assistants become a backdoor for accessing restricted information, creating security and compliance risks that would make enterprise adoption impossible. Dust's implementation of permission controls is one of the features that separates it from simpler RAG-based AI tools.

Conversation Management and Team Workspaces

Dust provides shared team workspaces where AI agents, conversations, and configurations are managed collaboratively. Team members can see shared agent definitions, contribute to agent improvement, and access conversation history. Conversations with agents are persistent, maintaining context across sessions for ongoing projects.

The workspace model supports different levels of access -- administrators can create and configure agents, while regular users can interact with deployed agents without modifying their configuration. This separation ensures that agents are maintained by knowledgeable team members while being accessible to everyone.

Pricing

Dust's pricing reflects its positioning as a team productivity platform rather than an individual AI tool. The Pro plan costs EUR 29 per user per month and includes custom AI assistants, all data source connectors, and access to GPT-4, Claude, and Mistral for teams of up to 50 users. Business and Enterprise plans offer custom pricing for larger organisations with unlimited users, advanced admin controls, SSO, SCIM provisioning, and dedicated infrastructure.

At EUR 29 per user per month, a 20-person team pays EUR 580 monthly. This represents genuine ROI only if Dust saves meaningful time per person. If each team member saves 2-3 hours per week by having AI agents that can quickly surface and synthesise company knowledge -- a plausible outcome for knowledge workers in information-dense roles -- the investment pays for itself relative to the loaded cost of employee time. For teams with simpler workflows or less distributed knowledge, the value proposition is harder to justify.

Our value assessment scores Dust 6.5 out of 10. The per-user pricing is expensive for small teams and there is no free tier for evaluation. A 14-day trial or a limited free tier would significantly lower the adoption barrier. The absence of a free plan means organisations must commit to a paid subscription before they can fully evaluate whether Dust's RAG capabilities improve their specific workflows with their specific data sources.

EU Compliance & Privacy

Dust earns a strong 8.5 out of 10 for EU compliance. As a French SAS headquartered in Paris, the company operates under EU jurisdiction and GDPR. Data processing occurs within the EU, and Dust provides a clear commitment that customer data is not used for model training.

The platform supports granular data access controls that mirror existing tool permissions, which aligns with GDPR's principle of data minimisation -- AI agents only process the minimum data necessary for an authorised user's query. Dust reports SOC 2 Type II certification, providing third-party assurance of security controls.

The compliance picture has a nuance: while Dust processes data in the EU, the underlying LLMs (GPT-4, Claude) are operated by US companies. Data sent to these models for inference may be processed in US infrastructure, depending on the model provider's configuration. Using Mistral as the LLM backend keeps the entire processing chain within French/EU infrastructure, which may be preferable for organisations with strict data residency requirements.

Who It's Best For

Knowledge-intensive teams of 10-50 people where information is distributed across multiple tools (Notion, Slack, GitHub, Google Drive) and employees spend significant time searching for context before doing their actual work.

Customer support and success teams that need AI agents grounded in product documentation, past tickets, and internal knowledge bases to draft accurate, contextual responses.

Engineering teams that want AI assistants capable of answering questions about codebases, technical documentation, and architectural decisions by referencing actual repositories and wikis.

European companies seeking a sovereign AI assistant that processes data in the EU and offers Mistral as an alternative to US-based LLM providers.

The Verdict

Dust is solving a real problem. Generic AI chatbots like ChatGPT are impressive but fundamentally limited by their inability to access your company's specific knowledge. Microsoft Copilot requires a full Microsoft 365 ecosystem commitment. Dust provides a model-agnostic, tool-agnostic AI assistant platform that works with the tools European teams already use.

The execution is strong: the agent builder is intuitive, the data connectors are reliable, and the permission-mirrored access is enterprise-ready. The main barriers are pricing (EUR 29/user/month with no free tier) and the dependency on data quality. Dust can only be as good as the knowledge it indexes. For well-organised teams with structured documentation, Dust delivers transformative productivity gains. For teams whose knowledge lives in unstructured Slack threads and outdated wikis, the investment in Dust should be paired with an investment in information organisation.

Frequently Asked Questions

Does Dust train AI models on my company's data?

No. Dust explicitly states that customer data is not used for model training. Your data is indexed for retrieval purposes only -- when an agent answers a query, it retrieves relevant documents from your connected tools and sends them as context to the LLM. Neither Dust nor the underlying model providers train on this data.

Can I use Dust with just one person?

Technically yes, but Dust is designed for teams. The platform's value increases with the number of connected data sources and team members contributing to shared knowledge. A single user with a single data source would get more value from a standard AI chat interface. Dust's strengths emerge when multiple people create and use specialised agents across multiple data sources.

How does Dust compare to Microsoft Copilot?

Both are AI assistants connected to company data. Copilot is deeply integrated with the Microsoft 365 ecosystem (Outlook, Teams, SharePoint, OneDrive) and requires Microsoft 365 licensing. Dust is tool-agnostic, connecting to Notion, Slack, GitHub, Google Drive, and others. Copilot is a better fit for Microsoft-centric organisations. Dust is better for teams using diverse tool stacks, particularly those common in startups and tech companies.

What happens if a data source connector breaks?

Dust monitors connector status and alerts administrators when data synchronisation fails. If a connector temporarily loses connection, previously indexed data remains available. The AI agent will note that its information may not be current. Persistent connector failures should be resolved through Dust's support channels or by re-authenticating the affected data source.

Can Dust be self-hosted?

As of 2026, Dust is available as a cloud service only, with data processing in the EU. Self-hosted deployment is not currently offered. For organisations with strict on-premise requirements, this is a limitation. The Enterprise plan offers dedicated infrastructure as a partial alternative, but the platform runs on Dust's managed cloud.