Contents
- Verdict
- What is Dust?
- Best for, and not for
- Strengths and limits
- Pricing
- What people actually build with Dust
- What Dust actually costs at small-business volume
- The gotcha: credits you cannot price, share or keep
- What breaks in production, and how you find out
- What it connects to that you already run
- Dust alternatives: what to pick instead, and when
- Migration and lock-in: can you get your work out?
- FAQ
Dust's seat prices are clear and its unit is not. Pro is €24 a seat a month billed yearly with 8,000 credits, and Max is €120 with 40,000, which is exactly five times the price for five times the credits: €0.003 a credit either way. So Max is not a discount, it is a bigger allowance. What nobody can tell you from the pricing page is how many credits a message costs, because that depends on the model, the context and the tools used. It is worth it for a team that wants shared agents over company knowledge with real admin controls. Be careful if your plan is automation through the API, which is a separate pool at $0.01 a credit with no free baseline.
What is Dust?
Dust is a workspace where a company's staff build and use AI agents that are connected to internal knowledge and tools. Agents are given instructions, data sources and tools, and employees talk to them in a shared interface or trigger them on a schedule or an event. Every plan includes more than 20 models from OpenAI, Anthropic, Google, Mistral and DeepSeek. Dust is sold per seat, and each seat comes with a monthly allowance of credits, the unit Dust uses to measure AI usage across messages, agent runs, tools and automated workflows.
Best for
Teams of up to about 100 people that want shared, governed agents over company knowledge without building anything themselves.
Not for
High-volume automated workloads driven through the API, which draw on a separately bought credit pool at a much higher rate.
Strengths
- All plans include 20+ models, and no model is locked to a higher tier
- Seat prices and credit allowances are published, including the monthly and yearly rates
- Admins get real spending controls: per-member limits, a shared automation cap and per-key caps
- A credit-spend checkpoint pauses a long task at 600 credits and asks before continuing
- US or EU data residency and a SOC 2 Type II listing on the pricing page
- Works with automation platforms including Zapier, Make, n8n and Power Automate
Limitations
- The pricing page gives no figure for how many credits a typical message costs
- Unused credits do not roll over, are not transferable between users and are not refundable
- Programmatic usage has no free baseline and is listed at $0.01 a credit on Business
- Business is limited to 3 data source connectors, 5 remote MCP servers and 5 Spaces
- Pay-as-you-go overage is Enterprise only; on Business a pool top-up must be bought in advance
- Enterprise pricing is not published
Dust pricing
Dust charges per seat, and each seat includes credits. A Free seat gets 500 credits once. Pro is €24 a seat a month billed yearly, or €30 billed monthly, with 8,000 credits a seat a month. Max is €120 yearly or €150 monthly with 40,000 credits. Prices exclude VAT. Credits reset each billing period and do not roll over. Programmatic usage through the API and triggers is listed at $0.01 a credit on Business. Enterprise is priced by sales with workspace-pooled credits.
| Plan | Price | What you get |
|---|---|---|
| Free | €0 | 500 credits, a one-time lifetime allocation that does not reset. No access to the workspace credit pool and no overage |
| Pro | €24/seat/mo | Billed yearly; €30 a seat billed monthly. 8,000 credits per seat per month. Excluding VAT |
| Max | €120/seat/mo | Billed yearly; €150 a seat billed monthly. 40,000 credits per seat per month. Excluding VAT |
| Business limits | Included | Applies to Pro and Max seats, for teams up to 100 people: up to 3 data source connectors, 5 remote MCP servers, 5 Spaces, email support, US or EU data residency |
| Programmatic usage | $0.01/credit | API calls, triggers and integrations draw only from the workspace credit pool, with no free baseline. Pool top-ups are at least 100 credits and valid for one year |
| Enterprise | Talk to sales | Workspace-pooled credits and volume pricing, unlimited connectors and MCP servers, SCIM, audit logs, custom data retention, single-tenant deployment, priority support with SLA |
Pricing reflects public plans as of November 17, 2026 and can change. Check Dust for the latest.
What people actually build with Dust
Dust is built around a simple idea: an agent is useful to a company when it knows what the company knows. So the builds that work are the ones where the hard part is finding and combining internal information, and a person is there to read the answer.
The builds that fit
A company knowledge assistant. An agent connected to the document store and the wiki answers questions such as what the refund policy is or where the onboarding checklist lives, with search and extraction across company data. This is the core use and the one the seat model is designed for.
A support drafting agent. Given the help centre and past tickets as data, it drafts a reply for a human to check and send. The person stays in the loop, which is where an agent over internal knowledge is safest.
A sales preparation agent. Before a call, it pulls what is known about the account from connected sources and the web and writes a one-page brief. Web search is a one-credit action, retrieval from company data a three-credit one.
A recurring report. Dust supports scheduled and event-driven triggers and multi-agent orchestration, so an agent can assemble a weekly summary without being asked. Note that triggers count as programmatic usage, which is billed from the workspace pool.
Inbox triage at our reference volume of 1,000 items a month, driven through the API or an automation platform. It works, and it is the expensive way to use Dust, for reasons the cost section sets out.
What it is genuinely bad at
High-volume, unattended automation. Dust's pricing is built for people having conversations, with a monthly allowance attached to each person. A process that fires thousands of times a month with nobody watching does not belong to a seat, so it draws on a pool you buy separately. It is also constrained on the Business tier for a company with scattered data: three connectors is not many once you count a document store, a chat tool and a CRM.
Bottom line: Dust fits interactive, knowledge-heavy work where an employee reads the answer, and fits poorly when the work is thousands of unattended runs through the API.
What Dust actually costs at small-business volume
The seat prices on dust.tt/home/pricing, read on 17 November 2026, are these. Pro is €24 a seat a month on yearly billing and €30 on monthly billing, with 8,000 credits a seat a month. Max is €120 a seat a month on yearly billing and €150 on monthly, with 40,000 credits. A Free seat has 500 credits as a one-time lifetime grant. All prices exclude VAT, and yearly billing is marked as saving 20 percent.
Divide price by credits and something useful appears. Pro yearly is €24 divided by 8,000, which is €0.003 a credit. Max yearly is €120 divided by 40,000, which is also €0.003. On monthly billing both are €0.00375. Max costs exactly five times Pro and includes exactly five times the credits. There is no volume discount between the two. You are not buying cheaper credits on Max; you are buying more of them for one person.
For a ten-person team on Pro, yearly billing is 10 times €24, so €240 a month, or €2,880 a year. Monthly billing is €300 a month. That team has 80,000 credits a month between them, but not shared: credits belong to each user, are not transferable, and do not roll over.
How far 8,000 credits goes is the question the pricing page does not answer. Its FAQ says credits are charged per message based on the model and the actions performed, and that a basic chat uses few credits while deep research uses more. It gives no number. The documentation gives the part that is fixed: action credits. Some tools are free. Basic actions such as web search or running another agent cost 1 credit. Advanced actions such as searching company knowledge, querying tables, image generation and third-party integrations such as Gmail, Slack or GitHub cost 3 credits. Token credits, which scale with the model and the length of the context, are added on top and are not published as a table.
| Plan | Price per seat per month | Credits per seat per month | Price per credit | 10 seats per month |
|---|---|---|---|---|
| Free | €0 | 500, once, lifetime | - | €0 |
| Pro, billed yearly | €24 | 8,000 | €0.003 | €240 |
| Pro, billed monthly | €30 | 8,000 | €0.00375 | €300 |
| Max, billed yearly | €120 | 40,000 | €0.003 | €1,200 |
| Max, billed monthly | €150 | 40,000 | €0.00375 | €1,500 |
| Programmatic usage (Business) | - | Bought into the workspace pool | $0.01 | 6,000 action credits = $60 |
| Enterprise | Talk to sales | Pooled across the workspace | Custom | Not published |
The reference workload through the API
Take 1,000 triage runs a month, triggered through the API. This is programmatic usage, which the documentation says draws only from the workspace credit pool and has no free baseline. The pricing page lists the programmatic usage rate on Business at $0.01 a credit.
We can price the action credits exactly. If each run does one company knowledge search (3 credits) and one third-party integration action (3 credits), that is 6 action credits a run, 6,000 a month, which at $0.01 is $60. Token credits are in addition and depend on the model and prompt length, so the true figure is higher than $60 by an amount we cannot calculate from published numbers.
Note the two currencies. Seat credits work out at €0.003 each and programmatic credits are listed at $0.01 each. They are different currencies and we have not converted them, but the gap is roughly threefold on the face of the numbers. An employee doing the same 1,000 tasks by hand inside their seat allowance would use credits that cost far less each than the automated version.
Bottom line: Dust Pro and Max both work out at €0.003 a credit on yearly billing, so Max is five times the allowance for five times the price, and API-driven usage is a separate pool at $0.01 a credit.
The gotcha: credits you cannot price, share or keep
Three properties of a Dust credit matter more than its price, and each is stated in the documentation.
The first is that you cannot work out in advance what a message costs. The total is token credits plus action credits. Action credits are fixed and published. Token credits scale with the AI processing involved, which means more capable models, longer contexts and more complex queries use more, and there is no published table. So two employees on the same plan can exhaust their allowance at very different speeds depending on which model their agent uses and how much company data it reads on each turn. The only way to know is to run your real agents for a couple of weeks and read the Members table, which shows credits consumed and remaining per person.
The second is that credits are personal. The documentation says they are not transferable between users, do not roll over to the next billing period and are not refundable. A ten-person team does not have 80,000 credits; it has ten allowances of 8,000. The heavy user runs out while the light user's credits expire unused. There is a workspace pool that anyone can be allowed to draw on after their own credits are gone, but on Business that pool is filled by top-ups the admin buys, in amounts of at least 100 credits, valid for a year.
The third is what happens at zero. A user is blocked when all available sources are exhausted or a limit is reached. On Business there is no pay-as-you-go: the documentation lists overage as Enterprise only. So a Pro user who runs out mid-month either waits, draws on a pool somebody topped up, or is upgraded. Dust offers an auto-upgrade setting that moves members to the next seat tier at their limit instead of blocking them, and its own documentation warns this may increase subscription costs. Going from Pro to Max is a fivefold increase for that seat, so leave that setting off until you understand your usage.
One protective feature is worth knowing. A task started from the web app pauses when it reaches 600 credits of cumulative AI spend and asks whether to continue. At €0.003 a credit that is €1.80 of allowance, and 8,000 divided by 600 is about thirteen such tasks in a month on Pro. Deep research and multi-agent tasks are how an allowance disappears, and the checkpoint is the warning. Admins can switch it off; do not.
Finally, backend functions in Frames cost 1 credit per second of runtime, so a function that runs for a minute is 60 credits before any tool calls.
Bottom line: Dust credits are personal, expire monthly and have no published per-message cost, so measure two weeks of real usage before choosing between Pro and Max for anyone.
What breaks in production, and how you find out
The most common failure is not technical. It is an employee being blocked. The documentation is specific about how it happens: limits are checked before each message starts, and a member is blocked when their seat credits and any pool access are used up, when a personal spend limit is reached, or when a group budget is spent. Dust reports the first applicable reason to the user. For the business, the symptom is somebody saying the assistant has stopped working in the third week of the month. Turn on the pool-threshold notification and review the Members table weekly at first.
The second is automation stopping. Programmatic usage, meaning API calls, triggers and integrations with tools such as Zapier, n8n and Google Sheets, draws only from the workspace pool. If the pool is empty on Business, that usage is blocked. A shared automation cap sets a monthly limit, with an email at 80 percent of the cap, and setting it to zero blocks all shared automations. Individual API keys can have a monthly credit cap too. These are good controls, and they mean an automated workflow can stop for a billing reason while every system reports itself healthy. Whatever triggers Dust should check the response and alert on a failure.
The third is the limit check itself. Because limits are checked before a message starts, the documentation notes a single message can take a member past their cap. A long multi-agent task started with a few credits left will run to completion, since an agent already generating a response finishes even if credits run out. That is the right behaviour for the user and a reason caps are approximate.
The fourth is the one every knowledge agent has: the answer is only as good as the connected data. An agent that confidently quotes a superseded policy does more harm than one that says it does not know. Decide which sources are authoritative, keep stale documents out of them, and give each source an owner.
Security and compliance, as published
The pricing page lists SOC 2 Type II and US or EU data residency. SSO with Okta, Entra ID or Jumpcloud appears in the Business feature list, with the comparison table noting it for 5 or more seats on demand. SCIM provisioning, audit logs, advanced security controls, custom data retention, single-tenant deployment and custom legal terms including a DPA are on Enterprise. The page does not mention HIPAA, so we are not stating a position on it. A business handling health data should ask Dust directly and get the answer in writing.
Bottom line: On Dust the production failure is a billing block, on a person or on an automation, so set the pool and automation alerts and have whatever calls the API notice when it is refused.
What it connects to that you already run
Dust's integrations come in three layers, and the Business plan limits two of them. The first is data source connectors. The pricing page lists connectors to more than 20 data sources and limits Business to up to 3, with Enterprise unlimited. Three is workable if your knowledge is concentrated, for example one document store, one chat tool and one ticketing system. It becomes the reason to talk to sales as soon as you need a fourth.
The second is tools through the Model Context Protocol. Business allows native MCP servers plus 5 remote ones, and Enterprise is unlimited. Third-party MCP integrations such as Gmail, Slack and GitHub, and all external MCP servers, are advanced actions at 3 credits a call. An agent that takes several such actions per message uses its allowance noticeably faster than one that only chats.
The third is the outside world calling Dust. There is a developer API, described as a conversation API and a data source API, and the pricing page lists automation platforms by name: Zapier, Make, n8n and Power Automate. This is the route for putting a Dust agent inside an existing workflow. Remember that it is programmatic usage and is billed from the workspace pool.
Inside the workspace, Spaces segment data and permissions, with 5 on Business. That is how you keep an HR agent's sources away from the sales team. Triggers can be scheduled or event-driven, and agents can orchestrate other agents, each sub-agent's cost accumulating into the total shown for the main one.
A sensible division for a small business is to use Dust for the interactive, knowledge-heavy work where staff benefit from an assistant that knows the company, and to keep high-volume deterministic automation on a platform that bills per execution or per task, calling a Dust agent only for the step that needs company knowledge.
Bottom line: Dust Business allows 3 data connectors, 5 remote MCP servers and 5 Spaces, and each third-party tool action an agent takes costs 3 credits.
Dust alternatives: what to pick instead, and when
The right alternative depends on which half of Dust you wanted. If you wanted an assistant layer for staff and your company runs on Microsoft 365, Copilot Studio builds agents where your documents and identity already are. If your knowledge is in a CRM, the agent attached to that CRM, such as HubSpot Breeze or Agentforce, starts with the data already connected.
If you wanted a self-serve no-code agent builder, MindStudio and Lindy are the self-serve builders to compare. Lindy in particular is also credit-metered, so the same advice applies: measure before you commit. Zapier Agents is the choice if the value is in reaching thousands of apps, with the connector catalogue doing the work.
If what you actually need is unattended automation at volume, with an AI step in the middle, a workflow platform is a better fit than any agent workspace. n8n bills a whole workflow run as one execution, and Make.com bills per step. And if you have developers and want agents in your own cloud with no usage meter from the vendor, Agno's runtime is free to serve with a flat fee for its control plane. Figures for the platforms we did not re-read today come from our August 2026 reviews; follow the links for the dated detail.
| Tool | Billing unit | Who it suits | Does better than Dust |
|---|---|---|---|
| Dust | Per seat, with a monthly credit allowance per person | Teams wanting shared agents over company knowledge | Model choice on every plan and detailed admin spend controls |
| Copilot Studio | Vendor's own usage meter (August review) | Microsoft 365 organisations | Sits inside the tools and identity a Microsoft shop already has |
| Lindy | Credits (August review) | Individuals and small teams wanting personal agents | Faster to a working personal assistant |
| MindStudio | See that review for dated figures | Teams wanting a self-serve no-code builder | A builder-first product; see that review |
| Zapier Agents | See that review for dated figures | Teams whose value is in app reach | The largest connector catalogue |
| n8n | Execution (whole workflow run) | Unattended automation at volume | Predictable cost for thousands of automated runs |
| Agno | Flat $150 a month control plane; serving is free | Python teams hosting agents themselves | No usage meter from the vendor |
Bottom line: Pick Dust for an interactive assistant layer over company knowledge, and a workflow platform that bills per run when the job is thousands of unattended automations.
Migration and lock-in: can you get your work out?
What you build in Dust is configuration: agent instructions, the selection of data sources, tool settings, Spaces and permissions. None of it is code you hold in a repository, so the first rule is to keep your own copy of the part that took thought. An agent's instructions are plain text. Keep them in a document outside Dust, with notes on which sources and tools each agent uses. With that, rebuilding an agent on another platform is an afternoon; without it, you are reconstructing it from memory.
Your data is in a better position than your agents. The knowledge Dust searches comes from your own systems through connectors, so the documents themselves never stop living in your document store, chat tool or CRM. Leaving Dust does not strand them. What you lose is the index and the conversation history, so export or save anything from past conversations that the business relies on.
Commercially, the choice between monthly and yearly billing is the lock-in decision. Yearly saves 20 percent and commits you for the year; monthly is €30 instead of €24 for Pro and can be stopped. For a first deployment, pay monthly for two or three months while you learn what your usage really is, then move to yearly once the seat mix is settled. The saving on ten Pro seats is €60 a month, which is small next to the cost of committing to the wrong tier.
Credits are the other thing you cannot take with you. Seat credits expire each billing period and are not refundable, and pool top-ups are valid for one year. Buy pool credits in small amounts matched to measured programmatic usage; do not buy a large balance in advance.
If you connect Dust to other tools through its API, keep that integration thin. A workflow that sends a question and receives an answer can be pointed at a different agent service with little work. One that depends on Dust-specific behaviour in many places cannot.
Bottom line: Your documents stay in your own systems, so leaving Dust costs you the agent configurations and unused credits: keep the instructions in your own files and pay monthly until usage is known.
Dust FAQ
How much does Dust cost?
Pro is €24 a seat a month billed yearly, or €30 billed monthly, with 8,000 credits per seat per month. Max is €120 yearly or €150 monthly with 40,000 credits. A Free seat has 500 lifetime credits. Prices exclude VAT, as published on 17 November 2026. Enterprise is priced by sales.
What is a Dust credit?
Dust's unit for measuring AI usage. Sending messages, running agents, using tools and running automated workflows all consume credits. The total for a message is token credits, which scale with the model and context, plus action credits, which are fixed per tool action.
How many credits does a message use?
Dust does not publish a figure. Its pricing page says a basic chat uses few credits and deep research uses more. The fixed part is action credits: 0 for free tools, 1 for basic actions such as web search, and 3 for advanced actions such as company knowledge search or third-party integrations.
Is Max better value than Pro?
Not per credit. Pro is €24 for 8,000 credits and Max is €120 for 40,000, both €0.003 a credit on yearly billing. Max is five times the allowance for five times the price, so it only makes sense for a person who would otherwise run out.
Do unused Dust credits roll over?
No. Seat credits reset each billing period, do not carry over, are not transferable between users and are not refundable. Free seat credits are a one-time grant that does not reset. Workspace pool top-ups are valid for one year.
What happens when a user runs out of credits?
They draw on the workspace credit pool if an admin allows it and it has credits. Otherwise they are blocked until the next period, or an admin upgrades the seat. Pay-as-you-go overage is available on Enterprise only. A response already being generated finishes.
How is API and automation usage billed?
As programmatic usage. It draws only from the workspace credit pool, has no free baseline, and is listed at $0.01 a credit on Business. Admins can set a shared automation cap and a monthly cap on each API key.
Which AI models does Dust include?
The pricing page says all plans include more than 20 models from OpenAI, Anthropic, Google, Mistral and DeepSeek, with no model locked to a higher tier. Some models consume more credits per message than others.
What are the limits on the Business plan?
Business covers Pro and Max seats for teams of up to 100 people and allows up to 3 data source connectors, 5 remote MCP servers and 5 Spaces, with email support. Enterprise removes those limits and adds SCIM, audit logs and single-tenant deployment.
Looking at alternatives? Dust gives your staff a place to build and talk to agents over company knowledge, and it works best when people use it interactively. Autonoly is a managed service for the unattended workflows that run without anyone at a keyboard. If you want a governed assistant layer for a team, Dust is a strong candidate. If you want a specific process automated end to end, a workspace of agents is not the same thing as a finished workflow. See the Autonoly review.
