Artificial intelligence
In the world of artificial intelligence, Dust AI Paris positions itself as an operating system for agents. Founded by Stanislas Polu and Gabriel Hubert, the French start-up aims to revolutionize work by offering a platform that makes AI operational and adaptable to business reality. Unlike generic assistants, Dust AI specializes in building personalized agents capable of understanding a company’s internal context, accessing its knowledge bases and performing actions on its behalf. The company describes its mission as a desire to transform the way work is done: “Our infrastructure connects models to enterprise data, transforming raw capabilities into agents that do real work”. This approach aims to create a system equivalent to what Windows was for microcomputing: a standardized ecosystem of AI “primitives” enabling custom agents to be built.
Do you have a question about the DUST.AI solution? Contact us now!
The Dust IA APP philosophy is based on several principles:
Making AI useful: the aim is not to create ever larger models, but tointegrate existing models into real workflows. The founders insist that value lies in the connection between AI and business processes.
Universality of primitives: just as Windows provided universal interface primitives for applications, Dust provides AI primitives (agents, connectors, tools) to make workflows intelligent.
Security and governance: the company focuses on security standards (SOC 2, HIPAA, RGPD) and data control. Protection of sensitive information and access control are an integral part of the platform.
Team spirit and speed of execution: internal principles evoke the importance of collaboration, rhythm and optimism. This culture influences product development, with priority given to rapid iteration and usability.
Dust AI enables each team to design agents adapted to its needs. The platform offers an engine that combines instructions, tools and access to internal data. The key points are:
Agent creation with no limits on complexity: users can develop agents for simple workflows or sophisticated enterprise integrations. Adjustable instructions and predefined templates speed up design.
Context integration: Dust connects to tools such as Notion, Slack, GitHub and websites. This enables agents to access collective knowledge and execute searches on these sources via APIs. This capability produces answers rooted in corporate data, unlike assistants that rely solely on public knowledge.
Multi-model and scalable: the platform is model agnostic, allowing you to switch between OpenAI, Anthropic, Gemini, Mistral and other models, so that you always use the most efficient. This flexibility ensures rapid adaptation to advances in LLMs.
Sharing and feedback: Dust encourages collaboration by enabling you to share agents, gather feedback and iterate quickly (dust.tt).
Extensibility: technical teams can code their own tools and integrations to meet specific requirements. Possibilities include adding custom connectors, semantic search, SQL queries or data visualizations.
Dust AI attaches great importance to safety:
Regulatory compliance: the platform is SOC 2 certified, HIPAA compliant and RGPD compliant. It allows you to select the hosting location to ensure data sovereignty, and offers fine-grained access controls.
Selection of trusted models: the company only works with models that meet strict security requirements.
Dust AI covers many fields. The platform offers agent templates for the following functions:
Sales: create customer snapshots from CRM history, generate call scripts, respond to RFPs, analyze conversations to improve sales performance.
Marketing: content writing in line with the charter, creation of product launch messages, translation adapted to the corporate tone and extraction of insights from feedback.
Customer support: contextual response generation, ticket synthesis, intelligent routing and feedback analysis for product improvement.
Engineering and data: data extraction from GitHub repositories, technical documentation generation, log analysis and test automation.
Legal, HR and finance: creation of agents capable of analyzing contracts, answering HR questions or generating financial summaries.
This diversity of applications explains Dust AI’s rapid growth, with over 1,000 corporate customers.
The case of Mirakl illustrates the value of Dust AI. Mirakl, a European pioneer of B2B marketplaces, had already invested in ChatGPT Enterprise in early 2023 and recorded a 55% utilization rate. However, the team led by Anne-Claire Baschet (Chief Data & AI Officer) found limitations: the inability to tap into internal documentation (Confluence), Zendesk data and Slack conversations prevented the creation of complex workflows.
After a year of experimentation, Mirakl has identified three key levers:
Limitless platform: we needed a tool capable of connecting all internal knowledge, using various business systems (Salesforce, Google Drive, etc.) and performing actions in these tools. The platform also had to be model agnostic, to adapt to future innovations.
Team skills development: Mirakl wanted to transform its users into agent builders thanks to an accessible interface and no-code templates to democratize AI.
Enterprise-class infrastructure: the organization handling sensitive data required regional hosting, no data retention and strict permission controls.
Adopting Dust AI enabled Mirakl to overcome these obstacles by offering a flexible, secure and agent-oriented platform. Thanks to connectors, teams were able to automate sales appointment preparation by retrieving information from Salesforce and Google Drive and generating personalized briefs. The company also saw an increase in employeeautonomy: more staff started to create their own agents, reinforcing the ownership of AI internally.
Developer-oriented platform: unlike purely no-code solutions, Dust offers declarative specifications, enabling versioned files to be used and agents to be managed like software projects. This approach is aimed at engineering teams looking for robustness and code control, while keeping templates accessible.
Deep connection to corporate knowledge: agents can leverage internal databases (Notion, Confluence, Zendesk, Slack, Google Drive) to provide contextualized answers, avoiding the “hallucination” that models not connected to internal data can do.
Multiple models and seamless changeover: the ability to switch easily from ChatGPT to Claude, Gemini or Mistral means you can constantly optimize agent performance.
Advanced security and governance: Dust strives to offer regional hosting, granular permissions and templates selected for compliance, meeting the requirements of large enterprises.
Extensibility: thanks to the ability to create customized connectors and tools, teams can adapt the platform to their unique processes.
Learning curve for non-developers: although Dust offers simplified templates and interfaces, its developer-first orientation may require technical skills to take full advantage of declarative specifications and custom integrations.
Ecosystem under construction: Dust is a relatively young company. Some integrations and templates are less numerous than with older players. However, the platform is growing fast and enriching itself regularly.
Customized pricing: our pricing policy is not public, and requires you to contact the sales team, which can be a deterrent for SMEs looking for transparent plans.
Dust AI aims to become theAI operating system for enterprises, connecting AI models to internal data and transforming employees into agent designers. Its strengths lie in integration depth, security, model flexibility and developer-first philosophy. Mirakl’s example shows that an agentic platform can help a company evolve from a simple adoption of generic tools to a genuine ecosystem of agents that automate complex tasks and generate a tangible return on investment. This makes Dust AI a key player for technophile companies looking for a secure, scalable and forward-looking solution.