Dust AI

Dust is an AI agent platform built for teams that want their AI systems to work with company knowledge, business tools and real workflows. Instead of limiting AI to a single chat window, Dust lets organizations create and share specialized agents, connect those agents to internal systems, choose different AI models, and coordinate work between people and agents.

That makes Dust a different proposition from general-purpose assistants such as ChatGPT or Claude. Those products can be used for business work, but Dust is designed around an organization’s own operating context: the data employees already use, the tools they already depend on, and the processes they want to automate.

Dust Is Built Around the Company, Not the Chat Window

The easiest way to understand Dust is to think of it as an orchestration layer between AI models and company systems.

An agent can be configured with instructions, knowledge and tools. That agent can then work with connected business information rather than producing answers from a blank conversation. Dust currently says it supports more than 70 connectors, including Slack, Notion, Google Drive, GitHub, Salesforce and Zendesk. Its public pricing page separately describes 20+ connector-supported data sources in the Business plan, so users should distinguish between the overall connector ecosystem and what a particular seat or plan can access.

This architecture matters because many business processes are scattered across multiple systems. A sales researcher may need CRM information, notes, emails and customer documentation. A support team may need ticket history, product documentation and internal policies. A data analyst may have to combine dashboards, definitions and operational records.

Dust’s goal is to make those disconnected sources usable by an agent in the same workflow.

The product also uses a shared workspace approach. Dust calls these collaborative environments Pods, where people and agents can work around a common project or topic. The platform describes this as “multiplayer AI,” meaning the AI is treated as part of a team workflow rather than a completely separate assistant.

One Workspace Can Use Many Different AI Models

Model choice is another important part of Dust.

The platform currently advertises access to more than 20 frontier and open-source models from providers including OpenAI, Anthropic, Google and Mistral, with DeepSeek also named on the pricing page. Teams can select a model for an agent and change that choice later without rebuilding the entire agent’s logic.

That model-agnostic approach changes the buying equation.

A company does not necessarily have to standardize every workflow around one model vendor. A research agent might use a model that performs well on long-form analysis, while another workflow could use a more economical model for repetitive classification or drafting.

There is a cost consideration, though. Dust uses credits rather than giving every workload an effectively unlimited pool of AI usage. Credit consumption depends on factors such as the model used, task complexity and the tools an agent calls. A simple conversation can consume relatively few credits, while a deep research workflow with multiple tool calls can use substantially more.

That means model flexibility is not only a technical feature. It is also part of cost management.

Agents Can Do More Than Answer Questions

Dust becomes more interesting when agents are connected to actions and triggers.

The platform supports multi-agent workflows and automation that can run on schedules or events. An agent can be equipped with reusable skills, knowledge and tools, allowing a repeatable process to be packaged and reused rather than reconstructed each time.

Dust introduced Skills as a way to package instructions, knowledge and tools so multiple agents can use the same expertise. The company describes this as a way to avoid duplicating instructions across agents and to make updates propagate to every agent using a particular skill.

This is useful in organizations where AI workflows are not isolated experiments.

Imagine a company has a standard process for reviewing sales opportunities. Instead of giving every employee a different prompt, the organization can create an agent with access to the relevant systems and use shared instructions or skills to make the process repeatable.

The same pattern can apply to support triage, account research, internal reporting, competitive research or documentation.

Dust also supports MCP, allowing organizations to connect external tools and proprietary systems through the Model Context Protocol. The platform additionally lists developer tools including a Conversation API and, on Enterprise, a Data Source API.

For technical teams, this makes Dust more than a point-and-click agent builder. It can serve as part of a broader application architecture.

The Real Value Appears in Cross-Functional Work

Dust’s strongest conceptual fit is work that crosses organizational boundaries.

A support process may start with a customer ticket, pull relevant account information from a CRM, consult product knowledge and then draft a response. A sales workflow can collect signals from Salesforce, product documentation, conversation intelligence tools and internal research before producing an account brief.

Dust itself highlights workflows covering engineering, customer support, sales, marketing and data analytics. Its examples include ticket triage, account research, proposal drafting, campaign brief generation, self-service data questions, pipeline monitoring and report drafting.

The important point is not that Dust has a button for each task. The platform’s model is compositional: organizations build agents with the instructions, knowledge and tools required for the particular job.

That makes the learning curve different from a conventional assistant.

A user asking ChatGPT a question generally needs to formulate the prompt. A Dust deployment can require someone to think about which data source the agent should access, which tools it can use, how permissions should work, what skills it needs and which model should run the workflow.

The second approach can be much more appropriate for repeatable organizational work, but it introduces operational responsibility.

Governance Is a Major Part of the Product

Enterprise AI becomes much harder when an assistant can read internal information or make changes in business systems.

Dust addresses this with administrative and security controls. Its current enterprise materials describe SOC 2 Type II certification, US and EU data residency, SSO, SCIM, role-based access controls, audit logs, custom retention policies and single-tenant deployment. The platform also describes a dual-layer permission model designed to separate what agents can access from which people are allowed to use them.

That distinction is important.

Giving an employee access to an agent does not automatically mean the agent should have unrestricted access to everything that employee might theoretically request. Dust’s permission model is intended to give administrators more control over data, tools, systems and agents.

The platform also provides usage analytics and cost visibility. Administrators can track usage, adoption and spending by agent and model, which becomes increasingly relevant when AI activity is measured through credits.

Dust says it does not train models on customer data and provides encryption at rest and in transit. Enterprise materials also describe HIPAA compliance enablement and GDPR compliance. Organizations with regulated or sensitive workloads should still verify contractual, residency and compliance requirements for their specific deployment rather than treating a platform-level claim as a substitute for their own review.

The New Pricing Model Changes the Buying Decision

Dust’s current public pricing is credit-based.

The Business plan offers Free, Pro and Max seat types. The Free seat includes 500 credits for its lifetime. Pro is listed at €30 per month, or €24 per month when billed yearly, with 8,000 credits per seat per month. Max is €150 per month, or €120 per month billed yearly, with 40,000 credits per seat per month. Enterprise uses custom pricing.

One detail is easy to miss: the plan is not simply a flat amount of AI access. Different actions can consume different amounts of credits, and unused monthly credits do not roll over. Dust says Pro and Max can continue through workspace overage when enabled by an administrator, while Free users are prompted to upgrade after their allowance is exhausted.

For that reason, teams should estimate real workflows before committing to a large rollout.

A team running lightweight internal assistants may have a very different consumption profile from one running deep research, code execution, data retrieval and multi-step automations every day.

The pricing model also makes the Free tier more useful as an evaluation environment than as a permanent high-volume solution. It provides a bounded amount of usage, allowing organizations to test the workflow before moving to a paid seat.

Where Dust Fits Compared With Other AI Platforms

Dust occupies a middle ground between enterprise AI assistants, automation platforms and custom agent infrastructure.

Microsoft Copilot Studio is closely tied to the Microsoft ecosystem and is a natural comparison for organizations already deep into Microsoft 365. ChatGPT Enterprise provides a broader general-purpose AI environment for employees. Glean emphasizes enterprise search and organizational knowledge. Relevance AI, Lindy and Stack AI address other forms of agent building and workflow automation.

Dust’s differentiator is the combination of model choice, company context, custom agents, connected tools, multi-agent workflows and governance inside one team-oriented platform.

The product’s current “multiplayer AI” positioning also matters. Dust is not merely asking employees to use a smarter chatbot. It is trying to create shared infrastructure where people build agents, agents use business systems, and teams improve the workflows over time.

That approach will not suit every organization.

Companies looking for a simple personal AI assistant may have little reason to introduce an enterprise agent platform. Teams without an owner for permissions, connectors, agent quality and credit usage may also struggle to get full value from the flexibility Dust provides.

For organizations with repetitive cross-system work, however, the proposition is much clearer. Dust provides a structured way to connect AI models to company context and then turn that context into reusable workflows.

The result is best understood not as “another AI chatbot,” but as infrastructure for deploying and operating AI agents inside a business.