Dust

Dust operates an enterprise AI agent platform that lets business teams build, deploy and manage agents connected to internal company knowledge and tools. The product centres on a shared collaborative workspace in which employees and agents work on the same projects, conversations, files, notifications and to-do lists, rather than each employee running an isolated assistant. The company describes the prevailing pattern of one assistant per employee as "single-player AI" and positions its own system as the "multiplayer" alternative, where context, tools and goals are shared and improvements to one agent propagate across the organisation.

Agents can analyse, transform and generate files including documents, spreadsheets, presentations and interactive data visualisations, and act across connected systems through a context layer that combines semantic search over company knowledge with integrations to more than 100 data sources and business tools, among them Slack, Notion, GitHub, Google Drive, Salesforce, Confluence, HubSpot, BigQuery, Snowflake and Zendesk. Built-in memory and feedback loops are intended to let agents learn from team preferences, usage patterns and feedback and to suggest improvements. Custom actions can be built as "Dust Apps", and the platform is model-agnostic, giving customers a choice of frontier models such as GPT-4 and Claude within a central governance layer. Enterprise features include granular permissions and private spaces, cost and usage monitoring, audit trails, agent analytics, SSO and SCIM provisioning, and regional hosting. The platform is SOC 2 Type II certified and GDPR compliant, offers EU and US data residency, and states that models are not trained on customer data under contractual commitments from major model providers. Dust's core repository is published on GitHub under an MIT licence.

Founding story

Co-founders Gabriel Hubert and Stanislas Polu met at Stanford in 2007 and previously built data analytics company TOTEMS, which they sold to Stripe in 2014. Polu subsequently worked as a research engineer at OpenAI on mathematical reasoning under Ilya Sutskever and co-authored papers on AI reasoning, while Hubert was chief product officer at French health-tech company Alan and had helped scale AI adoption at Stripe. They started Dust to move enterprise AI beyond per-person assistants toward shared, governed human-agent collaboration.

Business model

Dust sells software-as-a-service subscriptions to companies, which deploy agents across functions such as operations, support, marketing and sales. The company promotes an internal "AI Operator" role — non-engineering staff closest to the work who build and run agent fleets for their teams — as the intended adoption model, and says it runs primarily on its own product.

Published pricing lists a Pro tier at EUR 29 per user per month excluding tax, including advanced models, custom agents and actions, key connections, unlimited messages under fair use, fixed pricing on programmatic usage and up to 1 GB of data sources per user; an Enterprise tier for 100-plus users is custom-priced and adds SSO, larger storage, SCIM provisioning, regional hosting and priority support.

Traction

As of the May 2026 Series B announcement, Dust reported more than 3,000 customer organisations globally and more than 300,000 agents deployed on the platform, with monthly active adoption consistently above 90% and weekly active usage above 70% across customers, and zero customer churn in 2025 alongside 240% net revenue retention. Monthly active users were reported at 51,000 in the company's release and at 41,000 as of April 2026 in other coverage. The company crossed $20m in annual recurring revenue earlier in 2026, up from EUR 1m in ARR two years prior, and is not profitable. Customer-reported outcomes include a 46-person revenue team at Vanta estimating more than 400 hours saved weekly (with 900 people across sales, customer success and revenue operations cited in the company release), Clay growing its go-to-market team fourfold without a proportional rise in enablement headcount, Watershed reducing a recurring data-mapping workflow from two to three hours to a few minutes at a 78% success rate, and Qonto reporting roughly 50,000 hours saved annually across more than 50 specialised agents and over 1,000 daily users. Dust holds a 4.9 out of 5 rating on G2 from 19 reviews, and its GitHub repository has about 1,400 stars and 334 forks.

Latest developments

On 18 May 2026 Dust announced a $40m Series B co-led by Abstract and Sequoia with participation from Snowflake Ventures and Datadog, taking total funding above $60m. The company said proceeds will fund research and development and US growth, specifically agents that improve as they are used, collaboration primitives making humans and agents bidirectional co-contributors, and orchestration and governance infrastructure for enterprise scale. Run-rate revenue was not disclosed in the announcement.