Dust

Dust (dust.tt) is an AI assistants platform for teams that lets you build custom AI assistants connected to your actual knowledge sources, including Notion, Slack, GitHub, Google Drive, Confluence, and custom data via API. The product is built by a Paris-based team and has strong European enterprise adoption. Pricing starts at $29/user/month for Pro. The backend is open source under MIT license on GitHub, so teams with engineering capacity can self-host. Dust sits at the intersection of enterprise knowledge management and agent platforms, serving teams that want AI that actually knows their internal documentation and processes.

Dust launched out of Paris in late 2022, founded by a team with roots at Stripe and Alan. The core premise was that AI assistants were only as useful as the knowledge they had access to, and most AI products at the time gave you a capable model with no knowledge of your company at all. Dust's bet was that the hard part was not the model, it was the infrastructure to connect company knowledge to AI reliably and keep it current.

That bet has held up. In 2026, the gap between a generic AI assistant and one that actually knows your internal docs, your team's Slack discussions, and your codebase is the gap between a tool that helps occasionally and one that becomes a real part of daily work.

What Dust actually does

Dust is a platform for building and deploying AI assistants. The workflow is: connect your data sources, define an assistant (what it knows, what instructions it follows, what tools it can use), and deploy it to your team.

Data sources are the core. Dust connects to Notion, Slack, GitHub, Google Drive, Confluence, Intercom, and custom data via REST APIs. When you connect a source, Dust indexes it and keeps that index synchronized. Your Notion pages, Slack channels, GitHub repositories, and internal documents stay current in the assistant's knowledge base. This is not a one-time upload; it is a managed sync that updates as your content changes.

Assistants are built on top of those data sources with a system prompt and a model choice. A customer support assistant might have access to your product documentation in Notion, your historical support Slack channel, and your public knowledge base, with instructions to always answer in a specific tone and escalate certain question types. A developer assistant might have access to your GitHub repos, your architecture docs, and your internal engineering wiki, with instructions to always suggest tests alongside code changes.

The assistant model choice matters because Dust supports multiple providers. Claude, GPT-4o, and Mistral can each power different assistants in the same workspace. Teams that have developed opinions about which model handles which task types better can act on those opinions rather than being forced into a single provider.

Agent chaining for multi-step work

Beyond single-turn assistants, Dust supports agent chaining, where one assistant can call another or execute a sequence of steps. A research assistant might first query a Slack archive, then query a Notion knowledge base, then synthesize and format the result. A code review assistant might pull the diff from GitHub, check it against your style guide in Confluence, and produce a structured review.

This is more involved to set up than a simple assistant, but the capability matters for teams whose workflows are multi-step. Single-turn Q&A over a knowledge base is useful. Multi-step reasoning that combines information from different sources and takes a structured action at the end is substantially more powerful.

The agent chaining capability is one of the things that distinguishes Dust from simpler knowledge management tools. You are building workflows, not just chatbots.

Open source backend

The Dust backend is published at github.com/dust-tt/dust under the MIT license. This is a meaningful differentiator for certain customers.

For regulated industries, financial services, healthcare, or government, keeping all AI activity within your own infrastructure is sometimes required. The self-hosted Dust option provides a path to that. You deploy the platform on your own cloud account or on-premises environment, connect your data sources, and none of the data passes through Dust's servers.

The self-hosted path requires real engineering capacity. You need to deploy and maintain the application, manage the vector database for retrieval, and handle model API connections yourself. It is not a one-click deployment. But for organizations where the compliance requirement is genuine, the option exists and the codebase is actively maintained.

The pricing math

Pro at $29 per user per month is reasonable for teams with genuine knowledge management needs. For a 20-person team, that is $580 per month. The question is whether the productivity gains justify the cost, and the honest answer is that it depends heavily on how knowledge-dense your team's work is.

For teams where a significant portion of daily work involves finding or synthesizing internal information, the ROI case is straightforward. For teams with simpler information needs or fewer data sources worth connecting, the cost-to-value calculation is less clear.

The open source option is free for self-hosting, which matters for budget-constrained teams with engineering capacity. The cloud product's value is in the managed sync, hosting, and the ongoing product development that comes with being a paying customer.

Enterprise pricing is custom and adds SSO, SLA guarantees, dedicated support, and options for managed deployment in your cloud environment.

Competing with Glean and Notion AI

Glean is the comparison that comes up most often in enterprise evaluations. Glean is a broader enterprise search platform with more connectors, deeper IT-level admin controls, and a positioning as an org-wide search tool rather than a team-level assistant builder. Glean's pricing reflects this: it targets larger deployments at prices that are typically higher than Dust's per-seat rate.

Dust's angle is that team-level, task-specific assistants with curated knowledge are often more useful than org-wide search with everything in the index. A support assistant that knows exactly the documentation relevant to your product version is more useful to a support rep than a general search over the whole company's files.

Notion AI is another comparison point for teams already in Notion. Notion AI works within Notion's editing interface and is useful for writing and summarization within Notion. Dust is for deploying assistants to Slack and the web interface, grounding them across multiple sources beyond just Notion, and enabling multi-step agent workflows. If your knowledge lives only in Notion and your use case is writing help, Notion AI is sufficient. If you want to combine Notion with Slack, GitHub, and other sources in a deployable assistant, Dust is the more appropriate tool.

Setup and what to expect

Dust has a web onboarding that walks through connecting your first data sources. The most common starting point is Notion plus Slack, since those are the two most-connected sources and the ones where enterprise teams have the most institutional knowledge.

Building your first assistant takes longer than a simple prompt. You need to think about: what data sources this assistant should have access to, what instructions it should follow, what it should not do, and what model should power it. That design work is not burdensome once you have done it once, but it is real. Budget an hour or two to design and test a well-configured assistant versus a few minutes to get a generic chatbot running.

The quality difference between a thoughtfully configured Dust assistant and a generic AI chat is significant enough to justify that setup time if the use case is real. The teams that get the most out of Dust tend to be the ones that treat the assistant design seriously, not the ones that connect a data source and expect magic.

For teams with European data residency requirements, the Paris-based team and French enterprise customer base means Dust is more attuned to GDPR requirements and data residency concerns than many US-based competitors. This is not a feature that shows up in feature lists but it matters in procurement conversations in regulated European markets.

Key features

• Connect to Notion, Slack, GitHub, Google Drive, Confluence, and custom APIs
• Build custom AI assistants with specific knowledge, tools, and instructions
• Managed data sync keeps assistant context up to date as connected sources change
• Agent chaining for multi-step workflows across data sources
• Full audit log and permission controls for enterprise deployments
• Open source backend (MIT) available on GitHub at dust-tt/dust
• Works with GPT-4o, Claude, and Mistral models as backend options

Pros and cons

Who is Dust for?

• Engineering teams building internal tools assistants connected to GitHub and docs
• Customer success teams with an assistant that knows your product docs and Slack history
• Ops and finance teams querying internal process docs without bothering colleagues

Alternatives to Dust

If Dust isn't quite the right fit, the closest alternatives are glean , claude-app , perplexity , and lindy . See our full Dust alternatives page for side-by-side comparisons.