Dust is an enterprise AI agent platform built by Dust Labs, founded in 2022 by Gabriel Hubert and Stanislas Polu. The pair previously built TOTEMS, acquired by Stripe in 2014; Polu spent the intervening years as a research engineer at OpenAI. The company is headquartered in Paris and sells primarily to mid-size and enterprise technology teams.
The product's organizing idea is that agents should be shared infrastructure rather than personal chat sessions. A workspace connects company data sources — Slack, Notion, GitHub, Drive, CRMs and remote MCP servers — into a semantic context layer, and teams then build named agents on top of it that anyone in the workspace can invoke. Agents can be scheduled, triggered by events, and chained into multi-agent workflows. Model choice is per-agent across 20+ frontier models, which means a change in the underlying model market does not require rebuilding the stack.
Traction is real and unusually well documented. Dust raised a $40M Series B in May 2026 co-led by Sequoia and Abstract, with Snowflake Ventures and Datadog participating, bringing total funding above $60M. The company reports $20M ARR, 3,000+ organizations, 51,000 monthly active users, 300,000+ agents deployed and 240% net revenue retention, with named customers including Datadog, 1Password, Clay, Vanta and Doctolib.
The honest caveat is pricing. In June 2026 Dust replaced a flat per-seat plan carrying unlimited fair-use with a credit-metered model, and for heavy users that is a cost increase rather than a repackaging. Credits do not roll over month to month, and consumption depends on which model an agent uses and how many tool calls it makes — so a research-heavy workflow can drain a Pro seat well before the month ends and push a team toward the five-times-more-expensive Max tier. Teams should pilot a representative workload before committing to seat counts. One further note on the open-source claim: the core repository is genuinely MIT-licensed, but Dust is sold and operated as hosted SaaS, and self-hosting is not a marketed or supported path for most buyers.
Key Benefits
- No model lock-in: Per-agent model selection across five major labs means the platform survives shifts in which model is best for a given task.
- Agents as shared assets: Building an agent once and publishing it workspace-wide spreads value beyond the person who configured it.
- Grounded in real company data: The context layer syncs bi-directionally with connected sources, so agents work from current internal knowledge rather than a stale upload.
- Enterprise-grade controls: Dual-layer permissions, SCIM, RBAC and 365-day audit logs address the access questions that stall most internal AI rollouts.
- Open where it matters: An MIT-licensed core and a documented REST API give engineering teams inspection and extension paths that closed platforms do not.
Use Cases
- Internal knowledge assistants — Connect Notion, Slack and Drive so employees can ask questions against company documentation instead of interrupting colleagues.
- Sales and customer research — Build agents that pull CRM records and product usage into account briefs ahead of customer calls.
- Engineering workflow automation — Trigger agents on GitHub events to summarize pull requests, draft release notes or triage incoming issues.
- Scheduled reporting — Run multi-agent workflows on a schedule to assemble recurring analyses from several connected systems into a single digest.
