Dust: Product, team and how it was built
Dust (Platform to build and run AI agents on company data, France and United States): Used by more than 3000 organizations, May 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders at the start in 2023; 107 people listed on the about page in November 2026; venture funding in 2024 and 2026. Launched September 2022; still running.Added
- Outcome
- Used by more than 3000 organizations, May 2026
- Evidence
- E no revenue figure was checked
- Launch
- September 2022
- Status
- Still running
- Team
- 2 co-founders at the start in 2023; 107 people listed on the about page in November 2026; venture funding in 2024 and 2026
- Outside funding
- US$16M Series A, June 2024; US$40M Series B, May 2026; more than US$60M in total, per TheNextWeb
- Main channel
- Not stated
What is not verified
No revenue figure was found. TheNextWeb wrote on 18 May 2026 that run-rate revenue was not disclosed. The customer count, the 41,000 monthly active users and the zero churn in 2025 are statements of the company. A reader could look for a payment-processor profile, a filing or an interview that gives revenue. This story is grade E: the business is real and its product, price and customers are described from public pages, but no revenue figure was checked. Aiolum shows it as an idea worth studying, not as proof that it makes money.
Dust sells a platform on which companies build and run AI agents on their own data. An AI agent is a program that uses a language model, tools and data to do a task, not only to answer a question. The home page says that Dust "connects your company knowledge, tools, and teams so you can create, share, and run agents across real workflows". Customers choose among 20 or more models from providers such as OpenAI, Anthropic and Google.
Dust started with two co-founders and later grew; its about page lists 107 people in November 2026. TheNextWeb says Gabriel Hubert and Stanislas Polu founded Dust in February 2023, and the French register dates the company to that month. An Internet Archive capture of dust.tt from 14 September 2022 shows an earlier page that asks visitors to sign up "for when we're ready". TheNextWeb reports a US$16 million Series A in June 2024 and a US$40 million Series B on 18 May 2026.
No revenue figure was found, so this story has grade E. TheNextWeb wrote that run-rate revenue was not disclosed. The headline is the company's own count of organizations that use Dust, as TheNextWeb reports it on 18 May 2026. No downloaded source states that a named customer paid.
Why did Dust work?
Four causes, in the order they mattered. Each one is a reusable lesson.
The founders bet on the product layer above the models
Polu said on the Sequoia Capital podcast that the models were already useful and available in the API. He said the revenue of OpenAI was "ridiculously small" at the time. Hubert said the product layer for companies is "actually very thick". The archived 2022 page says the team takes a "pragmatic, product-driven approach" and is not training a new model.
Customers choose the model for each agent they build
The pricing page says that all plans include 20 or more models. It also says "You choose the model per agent." Hubert said on the podcast that the default models at first launch were OpenAI's. The company says Dust connects models from OpenAI, Anthropic, Google, Mistral and others.
Teams buy seats with credits, and large buyers talk to sales
The Business plan sells Free, Pro and Max seats with a monthly credit allowance. Enterprise is sold by contract with a sales conversation. The about page lists roles such as account executives and solution engineers.
How it was built
Few build details are published. These come from the company's own pages and a podcast, and are grade C. [8, 9, 4, 3]
- The 2022 page describes a product for "tinkerers" to design and deploy generative model apps.
- The June 2023 page says Dust brings internal data into context with "fully managed semantic search engines" and connects Notion and Slack as "continuously updated embeddings".
- The 2026 home page lists connectors for Slack, Notion, Google Drive, GitHub, Salesforce and Zendesk. It also says Dust has SOC 2 Type II certification and data residency in the US or EU.
- Hubert said on the podcast that the default models at first launch were OpenAI's. The home page links to a GitHub repository, which this page did not read.
Tools named in the sources
Each one opens its page in the Tools tab, with every story that uses it.
Who tried this before Dust, and why did they fail?
Three companies in our database of failed startups sold AI work tools or shared team knowledge to businesses. The values below are from that database and are not independently verified.
Olive AI
The promiseAn AI workforce that would automate the administrative work of hospitals and health systems.
Why it diedThe record says sales promised "AI workers" while delivery needed custom integrations for each client, and the bots broke when hospital software changed.
ROSS Intelligence
The promiseAn AI research platform that searches large legal databases for case law.
Why it diedThe record says it struggled against incumbents with decades of data, and investors were hesitant to keep funding a long sales cycle.
Memo
The promiseA place for engineering, product and design teams to create, share and archive rich media memos.
Why it diedThe record says it could not compete with entrenched incumbents that quickly added similar features.
The failed startups come from our database of 1,749 dead startups, retold here in our own words. Its entries are AI-assisted summaries of public sources, so these cards carry no evidence grade.
What changed
Each of the three sold work software to companies, and the record pages name the same two risks. Olive AI promised more automation than it delivered. ROSS Intelligence and Memo faced incumbents that held the data or added the feature.
Dust's pages show two differences. The customer chooses the model for each agent, and the company sells seats with a credit allowance instead of promising whole departments' work. Dust also connects to tools that a company already uses, such as Slack and Notion. Whether these differences avoid the same failures is not shown here.
Can one person copy Dust now?
Not for one person, today. A narrow version is open.
Dust grew from two founders to 107 listed people, according to the about page. TheNextWeb reports more than US$60 million of venture funding. A solo founder cannot copy that scale. A solo founder could copy one narrow part: agents for one department on one set of data.
Where it is still open: A single department or industry, where the data connectors and the rules for access are few.
- 1Pick one repeatable job in one department, such as sorting support tickets.
- 2Connect only the one or two tools that the team already uses.
- 3Let the customer choose the model, as the Dust pricing page describes.
- 4Charge per seat with a monthly credit allowance, and offer a free seat to try.
How much revenue does Dust make?
Each row shows the evidence grade and the source numbers from the list at the end of this page. The grades are explained in How Aiolum verifies.
| Fact | Value | Grade | How we know | Sources |
|---|---|---|---|---|
| Organizations that use Dust, May 2026 | more than 3,000 | C | TheNextWeb, 18 May 2026: "On the company's own numbers", Dust is "used by more than 3,000 organisations". SiliconANGLE reports the same count. The count does not separate paying organizations from free ones. The home page says "3,000+" with no source. | [1, 2, 4] |
| Monthly active users, April 2026 | 41,000 | C | TheNextWeb, 18 May 2026, on the company's own numbers. The same article gives 300,000 agents deployed and zero customer churn in 2025, also the company's statements. | [1] |
| Revenue | no figure found | E | TheNextWeb: "Run-rate revenue was not disclosed." No page downloaded for this story gives a revenue figure. | [1] |
| Funding | US$16M Series A, June 2024; US$40M Series B, 18 May 2026; more than US$60M in total | C | TheNextWeb, 18 May 2026: the Series B is co-led by Abstract and Sequoia, with Snowflake and Datadog. SiliconANGLE gives the same US$40M and total of more than US$60M. | [1, 2] |
| Founding of the company | 17 February 2023, as Permutation Labs in Paris; founders Gabriel Hubert and Stanislas Polu | C | The French company register gives 17 February 2023 as the creation date of Permutation Labs. The privacy page names Permutation Labs, Paris, as "Dust". TheNextWeb states February 2023 and the two founders. | [12, 11, 1, 3] |
| First dated trace of the product | 14 September 2022 | C | An Internet Archive capture of dust.tt titled "Dust - Build Generative Models Apps". It asks visitors to sign up "for when we're ready" and names no person. Its "Learn more" link goes to a Notion page titled "Dust" at the address spolu.notion.site. This is the earliest capture of the domain in the Internet Archive. | [8] |
| Site in June 2023 | sign-in with Google or GitHub; footer "Dust © 2022-2023" | C | An Internet Archive capture of dust.tt from 2 June 2023. It describes connecting Notion and Slack as "continuously updated embeddings". | [9] |
| Team at the start | 2 co-founders | C | TheNextWeb and the Sequoia Capital page name two founders. No source gives a headcount for 2022 or 2023. | [1, 3] |
| Team today | 107 name-and-role entries on the about page | C | Counted on 12 November 2026 on the company's about page, which includes the two co-founders and roles such as engineering, sales and operations. | [6] |
| Prices, November 2026 | Pro EUR 24 per seat a month billed yearly, EUR 30 monthly; Max EUR 120 or EUR 150 | C | The company's pricing page, excluding VAT. The Free seat has 500 credits for its lifetime. Enterprise is "Talk to sales". | [5] |
| How billing works | seats prorated mid-cycle and invoiced; unused credits do not roll over | C | The pricing page FAQ: a member added mid-cycle is "charged only for the remaining days" on the next invoice. | [5] |
| Time from the first trace to the May 2026 count | 44 months | calc | Calculated: September 2022 to May 2026. | [8, 1] |
How to read this table. Grade C means a statement by the company, or a press article that repeats it. Grade E is the status of the whole story: no revenue figure. A "calc" value was computed on this page from the rows above it.
| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free seat | 0 | 0 | 500 credits for the lifetime of the seat |
| Pro seat | 24 | 30 | 8,000 credits a month; priced per seat |
| Max seat | 120 | 150 | 40,000 credits a month; priced per seat |
| Enterprise | talk to sales |
The Business plan is for teams of up to 100 people and has three seat types. Credits measure AI usage and reset each billing period. Enterprise has no published price, and the page offers "Talk to sales". A seat added in the middle of a billing period is prorated.
When did Dust launch, and how did it grow?
From a sign-up page in September 2022 to a Series B in May 2026. Each entry carries its grade.
- 14 September 2022CAn archived dust.tt page asks visitors to sign up "for when we're ready". It describes a product for tinkerers who design generative model apps. [8]
- 17 February 2023CPermutation Labs, the company behind Dust, is created in Paris, according to the French company register. TheNextWeb says Gabriel Hubert and Stanislas Polu founded Dust that month. [12, 1, 3]
- 2 June 2023CThe archived home page offers sign-in with Google or GitHub and describes semantic search over Notion and Slack data. [9]
- June 2024CA Series A of US$16 million, led by Sequoia, according to TheNextWeb. [1]
- 18 May 2026CA Series B of US$40 million is announced. The company says it has more than 3,000 organizations and 41,000 monthly active users in April. [1, 2]
- 12 November 2026CThe about page lists 107 people. The pricing page lists seats from EUR 0 to EUR 150 a month and an Enterprise plan by contract. [6, 5]
What could go wrong for Dust?
The risks are the mirror image of the causes above.
What is not verified
No revenue figure was found. TheNextWeb wrote on 18 May 2026 that run-rate revenue was not disclosed. The customer count, the 41,000 monthly active users and the zero churn in 2025 are statements of the company. A reader could look for a payment-processor profile, a filing or an interview that gives revenue.
No source states that a customer paid
The pricing page lists self-serve seat prices, a free start and billing terms, which is what the casebook (entry 33) asks for. TheNextWeb names Vanta, Watershed and Qonto from customer data points and case studies. SiliconANGLE quotes a partner of the investor Abstract about Datadog and 1Password. All of these repeat material of the company or of its investors.
Counters, logos and quotations have no source
The home page shows "3,000+", "300,000+" agents deployed, "Live Active right now", about 14 "Case study" cards and customer logos. It also shows quotations from three people: a Head of GTM Engineering at Clay, a Founder in Residence at Vanta and the CEO of Assembled. A search for one quoted sentence found only the company's page. The home page says "70+ connectors", the pricing page says "20+ data sources" for the Business plan, and TheNextWeb says "over 100 data sources".
Start date, entity and the earlier page
The product is first seen in September 2022, five months before the company was registered. The French register dates Permutation Labs to 17 February 2023, and TheNextWeb and Sequoia Capital say early 2023. The 2022 page names no person. Its link to a Notion page at spolu.notion.site and the 2023 footer "Dust © 2022-2023" suggest the same project. Sequoia's host says Polu was at OpenAI "from 2019 through late 2022". Sequoia is also an investor in Dust.
Sources
The grade next to each source is the best grade that the source can support. "Read in full" means that the text of the page was downloaded and the numbers were checked in it.
- C TheNextWeb, 18 May 2026: Dust raises $40M to push enterprise AI past the single-player eraRead in full. Source for the funding, the founding month, the founders and the company's own usage counts. It says run-rate revenue was not disclosed.
- C SiliconANGLE, 18 May 2026: Multiplayer AI startup Dust raises $40M to help enterprises move beyond isolated AI assistantsRead in full. Confirms the round, the total funding and the customer count. It repeats the company's statements.
- C Sequoia Capital, Training Data podcast: Dust's Gabriel Hubert and Stanislas Polu: Getting the Most From AI With Multiple Custom AgentsRead the summary and the transcript. The page has no date; the talk refers to 2024. Sequoia is an investor in Dust. Source for the founders' words and for "Founded in early 2023".
- C Dust home page, read 12 November 2026Read in full. The Forbes and Axios headlines it shows were not read: Forbes disallows Claude agents and Axios answered 403. The terms page answered a bot challenge and was not read.
- C Dust pricing page, read 12 November 2026Read in full. Source for the seat prices in euros, the credits and the billing terms.
- C Dust about page, read 12 November 2026Read in full. The entries with a name and a role were counted. It gives no founding year.
- C Dust blog index, read 12 November 2026Read the index. It shows posts and partnerships from October and November 2026 and nothing aimed at children or students.
- C Internet Archive capture of dust.tt, 14 September 2022Read in full. A short page titled "Dust - Build Generative Models Apps" with a sign-up form. It names no person. The Notion page behind its "Learn more" link was also read in its archived copy of the same day.
- C Internet Archive capture of dust.tt, 2 June 2023Read in full. Home page with sign-in and a footer "Dust © 2022-2023".
- C Database of failed startups, local exportRecord pages read on the website on 12 November 2026. AI-written summaries, not independently verified.
- C Dust privacy policy, read 12 November 2026Read the opening and the section on children. It names the company as Permutation Labs, Paris.
- C Annuaire des Entreprises (French state company register), search for Permutation Labs, read 12 November 2026Read the one result. It gives the creation date 17 February 2023 and a head office in the 4th arrondissement of Paris.
Run this company? Send corrections or a verified figure to aiolum@livyden.com