Dust

Imagine a complete team of employees being able to build digital assistants tailored to their daily work, without writing a single line of code, and with direct connection to the company’s databases and the tools they actually use every day like Notion, Slack, and Google Drive. This is exactly what Dust offers, a French-origin platform that has transformed from just a regular chat interface into a complete environment for building deeply customizable AI Agents, primarily aimed at companies and teams looking to leverage large language models within the context of their own internal data, rather than in a general void like traditional chatbots do.

In this review, we will dive into the practical details of Dust: how it works, what distinguishes it from ChatGPT for business, Glean, or Notion AI, where its true strengths lie, and where it shows its limits, with practical examples that any technical, marketing, or legal team can immediately benefit from.

What is the Dust Tool?

Dust (available at https://dust.tt) is a platform for building and deploying “AI Agents” tailored for enterprises. Rather than dealing directly with a single language model, Dust allows you to assemble a smart agent that combines: a language model of your choice (OpenAI’s GPT-4o, Anthropic’s Claude 3.5 Sonnet, Mistral Large, or Google’s Gemini), a collection of “Connections” to company data sources, and customized System Instructions that precisely define the agent’s personality and mission.

The core idea behind Dust is the concept of “Data Sources” and “Connectors,” where the platform continuously and simultaneously indexes content from Notion, Slack, Google Drive, Confluence, GitHub, Intercom, and Zendesk, and then uses Retrieval Augmented Generation (RAG) technology to feed the language model with the correct context from these sources when responding to any inquiry. This means that the agent you build does not “guess” answers, but actually searches within your company’s real documents before responding.

Key Features of Dust

• Multi-Agent Builder: Any team member can create a new agent via a simple interface, specifying its name, image, text instructions, and the tools it can use like web search or calling an external API.
• Multi-Model Support: Dust does not lock you into a single model; you can switch between GPT-4o, Claude, and Mistral depending on the task, providing great flexibility in managing cost and accuracy.
• Ready-made Connectors: Direct and ready-to-use connections with Notion, Slack, Google Drive, GitHub, Confluence, Intercom, Zendesk, and Websites via Crawling, eliminating the need to build custom integrations from scratch.
• Dust Apps (formerly): An environment for building advanced Pipelines that combine multiple language model calls, programming tools, and logical conditions, for those needing deeper customization than just a chat agent.
• Shared Workspaces: Every agent is built within a workspace that can be shared with the team, with access permissions controllable by roles.
• Chat History and Feedback: The capability to track every interaction with the agent and review the responses to improve the system instructions over time.
• Open API for Developers: The ability to call agents built in Dust from external applications via REST API, allowing their integration into your company’s products.
• Instant Context Addition (@mentions): Within the chat interface, you can invoke another agent or a specific data source using the “@” tag to direct the question more accurately.

How to Use Dust Step by Step

1. Create an Account: Go to https://dust.tt and register via email or Google account, then create a new Workspace named after your company or team.
2. Link Data Sources: From the “Connections” menu, activate the connection to Notion or Slack or Google Drive via OAuth, and grant the necessary permissions for indexing. The initial process can take minutes to hours depending on the amount of data.
3. Create a New Agent: Click on “New Agent,” choose a clear name such as “Technical Support Assistant,” and write the system instructions precisely, such as: “You are an assistant that answers only based on knowledge base articles in Notion, and if you can’t find the answer, request the user to contact the support team.”
4. Link Data Sources to the Agent: Specify which of the previously indexed data sources this agent should use when searching for context.
5. Select the Language Model: Choose GPT-4o for precise and complex tasks, or Mistral for lighter and faster tasks in terms of cost.
6. Test within the Interface: Try the agent with real questions from the team and review the accuracy of the citations it presents for each answer.
7. Publish and Share: Share the agent’s link with the team, or activate it directly within Slack via the Dust app for Slack, or call it from an external system via API.

Practical Benefits and Advantages

The benefit of Dust varies based on the type of team using it:

• Customer Support Teams: Build an agent that answers customer inquiries based on actual Zendesk and Intercom articles, reducing response time and maintaining accuracy instead of general answers that may be incorrect.
• Product and Engineering Teams: Link the agent to GitHub repositories to summarize Pull Requests or answer questions about internal code documentation without needing to manually search through dozens of files.
• Human Resources: An agent that answers employees’ questions about internal policies by referring to the HR-specific Notion pages, saving the HR team time on frequently asked questions.
• Sales Teams: Summarize client meetings or craft email replies based on stored product information internally.
• Collective Time Savings: Instead of each employee asking a colleague about repeated information, the agent becomes an immediate reference available around the clock via Slack.

Disadvantages and Challenges

Despite the strength of the idea, there are points that should be honestly considered:

• Initial Learning Curve: Understanding the difference between “Agent,” “Data Source,” and “Connection” requires time for the tech team to properly configure the initial setup, especially for non-technical teams.
• Dependence on the Quality of Source Data: If Notion or Confluence documents are poorly organized or outdated, the accuracy of the agent’s answers declines directly, as Dust does not create new knowledge but only retrieves existing one.
• Cost at Intensive Use: Calling models like GPT-4o intensively via Dust translates into additional billing based on the volume of usage, which large teams need to monitor carefully.
• Limitations of Some Connectors: Not all work tools are fully supported; some legacy systems require custom integration via API, which is not immediate.
• Lack of Strong Independent Mobile Interface: Current optimal use is through a browser or Slack, which may not suit teams relying on custom mobile applications.

Comparing Dust to Competing Tools

Dust vs ChatGPT Enterprise

ChatGPT Enterprise provides the power of the GPT-4 model with higher data privacy, but it is less specialized in building multiple agents linked to precise data from various sources as easily as Dust offers through its ready connectors. Dust gives you the freedom to choose the model while ChatGPT Enterprise is tied only to OpenAI models.

Dust vs Glean

Glean excels in comprehensive enterprise search across a larger number of sources with a deeper focus on rapid indexing for large organizations, while Dust stands out in the flexibility of building custom agents for specific tasks (like a separate technical support agent from a sales agent) more easily and quickly for small to medium teams.

Dust vs Notion AI

Notion AI is deeply integrated within the Notion space itself and suitable for those who primarily work within it, while Dust exceeds the boundaries of a single tool and combines context from multiple tools into one agent, which Notion AI does not provide due to its relatively closed nature.

Dust vs Microsoft Copilot

Copilot excels in deep integration with the Microsoft 365 environment (Outlook, Teams, Word), but Dust is more open to tools outside of the Microsoft ecosystem such as Slack, GitHub, and Notion, and more flexible in controlling each agent’s instructions independently.

Practical Examples of Using Dust

• Meeting Summary Agent: Linking the agent to transcriptions of meetings stored in Google Drive, directing it with instructions to extract only decisions and tasks from each transcription upon request.
• Code Review Agent: Linking it to a GitHub repository and directing it to summarize changes in a specific Pull Request and determine whether they comply with internally documented coding standards.
• Agent for Responding to Frequent Customer Inquiries: Linking it to the Zendesk knowledge base, and directing it to respond in a formal language with the source article link attached to each answer.
• Market News Summary Agent: Using web crawling to feed the agent with content from specific competitor websites and summarize any new updates weekly.
• Internal Assistant Agent via Slack: Activating the agent within a dedicated HR team Slack channel for immediate responses to employees’ questions about leave policy.

Pricing

Dust offers several pricing plans primarily aimed at teams and companies:

• Limited Free Trial: Initial trial with a limited number of messages and agents to test the platform before subscription.
• Pro Plan: Monthly subscription per user granting access to a larger number of agents and connectors, billed additionally to the cost of language model usage based on actual consumption.
• Enterprise Plan: Customized pricing including advanced security features, direct technical support, and dedicated SLA for large companies, requiring direct communication with the sales team through the official website.

It is always recommended to review the official pricing page at https://dust.tt for accurate and updated figures, as these plans are subject to periodic changes.

Evaluation and Getting Started Tips

Dust is ideally suited for technical and product teams in startups and medium-sized companies that already have organized data in tools such as Notion, Slack, and GitHub, and want to leverage this data into functional AI agents rapidly without building RAG infrastructure from scratch. It also suits customer support teams looking to ease the burden on human employees for frequently asked questions.

Conversely, Dust is not the best option for independent individuals or very small teams that do not have internal data sources that need indexing; they may find regular ChatGPT or Claude sufficient for their needs without additional complexity.

A tip for getting started: Don’t start by building ten agents at once. Begin with one agent with a clearly defined task (like answering only HR questions), test it for two weeks with a small team, monitor the accuracy of its answers, and then gradually expand to other agents after refining the system instructions based on actual feedback.

Conclusion

Dust represents a strong and practical option for any company looking to shift from using AI as a general tool to building dedicated AI agents that understand the internal work context accurately, thanks to its ready connectors with Notion, Slack, GitHub, and others, and its support for multiple language models that provide real flexibility in managing cost and quality. Its main challenges lie in the complete dependence on the quality of source data and the initial learning curve for non-technical teams, yet these challenges become secondary to the practical value it offers for support, product, and HR teams. For those seeking a tool that balances technical flexibility and ease of no-code building, Dust deserves a real trial through its official website https://dust.tt before making a decision to subscribe to an extended plan.

Do you make Dust? Add the DaleelAI badge to your site

Copy the code and paste it into your site’s footer or press page. It links to this page, so visitors can read about your tool in Arabic and English.

<a href="https://daleelai.com/en/dust-en/" target="_blank" rel="noopener"><img src="https://daleelai.com/wp-content/themes/DaleelAI/assets/badges/featured-light.svg" alt="Dust on DaleelAI — the Arabic AI tools directory" width="276" height="56" style="border:0"></a>Copy code

<a href="https://daleelai.com/en/dust-en/" target="_blank" rel="noopener"><img src="https://daleelai.com/wp-content/themes/DaleelAI/assets/badges/featured-dark.svg" alt="Dust on DaleelAI — the Arabic AI tools directory" width="276" height="56" style="border:0"></a>Copy code