Key takeaways
- Raised a $40M Series B (May 2026) led by Abstract and Sequoia with Snowflake Ventures and Datadog, bringing total funding past $60M — backed by 240% net revenue retention and zero customer churn in 2025
- "Multiplayer AI" positioning: instead of one assistant per person, teams build and share custom agents on company data, with 3,000+ organizations running 300K+ deployed agents
- Model-agnostic by design — users pick GPT-5, Claude, Gemini, or Mistral per agent — and surface-agnostic, reachable from Slack, web workspace, Zendesk, Chrome extension, API, and Zapier
FAQ
What is Dust?
Dust is an enterprise platform where teams build, share, and deploy custom AI agents connected to company data, callable from Slack, a web workspace, browser extension, and APIs.
How much does Dust cost?
Pro is €29 per user per month with a 14-day free trial; Enterprise is custom-priced with a 100-seat minimum and adds SSO, SCIM, and US/EU data hosting options.
What models does Dust use?
Dust is model-agnostic — agents can run on GPT-5, Claude, Gemini, Mistral, and other frontier models, selectable per agent within a shared governance layer.
How is Dust different from Glean?
Glean leads with enterprise search over company knowledge; Dust leads with building and sharing action-oriented custom agents, with search as one capability among many.
Executive Summary
Dust is a "multiplayer AI" platform: rather than giving each employee an isolated assistant, it gives the company a shared workspace where teams build custom agents on company data and tools, then deploy them anywhere work happens — Slack, a web workspace, browser extension, or API.[1] The company argues the hard part of enterprise AI is no longer prompting for output but "designing the system that allows humans and agents to share context, use the right tools securely, and facilitate reviews and handoffs."[1]
The thesis is working commercially. More than 3,000 organizations run over 300,000 deployed agents on Dust, and the company reported 240% net revenue retention with zero customer churn in 2025.[2] In May 2026 it raised a $40M Series B led by Abstract and Sequoia, with strategic participation from Snowflake Ventures and Datadog, bringing total funding past $60M.[3]
| Attribute | Value |
|---|---|
| Company | Permutation Labs SAS (Dust)[3] |
| Founded | 2022[4] |
| Founders | Gabriel Hubert (CEO, ex-Stripe), Stanislas Polu (ex-OpenAI research engineer)[3] |
| Funding | $40M Series B (May 2026); $60M+ total[3] |
| Investors | Abstract, Sequoia, Snowflake Ventures, Datadog[1] |
| GitHub Stars | ~1,380 (as of June 2026)[4] |
| Headquarters | Paris, France[3] |
Product Overview
Dust lets teams create custom agents — each with its own instructions, model, data sources, and tools — and share them across the organization. Agents connect to company knowledge (Notion, Google Drive, GitHub, Slack, Salesforce, and 100+ enterprise data platforms) and execute actions, not just answer questions.[3] Customers like Persona run 300 agents across 11 departments; Doctolib deploys Dust to 3,000 employees.[1]
The platform targets "AI Operators" — functional experts in RevOps, GTM engineering, product marketing — who build agents for their teams rather than waiting on engineering.[1] Usage depth backs the shared-agent model: weekly active usage exceeds 70% across the customer base.[3]
Key Capabilities
| Capability | Description |
|---|---|
| Custom Agent Builder | Define instructions, model, data sources, and tools per agent; share workspace-wide[5] |
| Company Data Connections | Notion, Google Drive, GitHub, Slack, Salesforce, and 100+ enterprise data platforms[3] |
| Action Execution | Agents act on connected tools, process files, and generate documents with cloud-hosted compute[3] |
| Multi-Model Selection | GPT-5, Claude, Gemini, Mistral — selectable per agent within a governance layer[6] |
| Shared Collaboration Workspace | Humans and agents coexist in projects with shared conversations and artifacts[3] |
| Programmatic Access | API, Google Sheets, and Zapier usage with included credits[6] |
Product Surfaces
| Surface | Description | Availability |
|---|---|---|
| Slack | Native integration — call shared agents from channels | GA[6] |
| Web Workspace | Primary surface for building and conversing with agents | GA[5] |
| Zendesk | Native integration for support workflows | GA[6] |
| Chrome Extension | Agents available in the browser | GA[6] |
| API / Zapier / GSheet | Programmatic agent invocation | GA[6] |
Technical Architecture
Dust is a hosted SaaS with a model-agnostic core: customers select preferred frontier models, which plug into Dust's governance layer for data access and tool permissions.[3] An intelligence layer connects agents to 100+ enterprise data platforms, and cloud-hosted compute handles file processing and document generation with integrated memory loops.[3]
The platform's code is open source under MIT (dust-tt/dust, ~1,380 stars, active daily commits), though the hosted SaaS is the commercial product rather than self-hosting being a supported path.[4]
Key Technical Details
| Aspect | Detail |
|---|---|
| Deployment | Hosted SaaS; US/EU data hosting options on Enterprise[6] |
| Model(s) | Multi-model: GPT-5, Claude, Gemini, Mistral — user-selectable[6] |
| Integrations | Slack, Notion, Google Drive, GitHub, Zendesk, Salesforce, 100+ data platforms[3] |
| Security | SOC 2, zero data retention; SSO (Okta, Entra ID, JumpCloud) and SCIM on Enterprise[6] |
| Open Source | Yes (MIT) — but SaaS is the product[4] |
Strengths
- Best-in-class retention economics — 240% net revenue retention and zero churn in 2025 indicate customers expand aggressively once deployed; few AI platforms publish numbers this strong[2]
- Shared agents compound — One operator builds an agent, the whole team benefits; Persona's 300 agents across 11 departments show the multiplayer model scaling inside a single customer[1]
- Model-agnostic hedge — Per-agent choice of GPT-5, Claude, Gemini, or Mistral means no single-vendor lock-in and instant access to whichever frontier model leads[6]
- Strategic investor signal — Snowflake Ventures and Datadog participating alongside Sequoia and Abstract suggests deep integration paths into enterprise data and observability stacks[3]
- Genuine usage depth — 70%+ weekly active usage is unusually high for enterprise AI tooling, where shelfware is the norm[3]
- Accessible entry point — €29/user/month with a 14-day trial and a 1-user minimum lets teams start small before an enterprise commitment[6]
Cautions
- Builder-dependent value — The platform assumes someone (an "AI Operator") will invest in building good agents; teams wanting an out-of-the-box assistant may stall before reaching value[1]
- Enterprise features gated at 100 seats — SSO, SCIM, US/EU hosting choice, and advanced security controls require the Enterprise plan with a 100-member minimum[6]
- Programmatic usage costs — Pro includes only "free credits" for API/Zapier/GSheet usage with fixed pricing beyond; heavy automation workloads add cost on top of seats[6]
- Open source in name more than practice — The MIT repo exists, but Dust does not position self-hosting as a product path; buyers should treat it as SaaS-only[4]
- EU-priced, EU-headquartered — Pricing in euros and a Paris HQ are neutral-to-positive for European buyers but US procurement teams should confirm support and data-residency specifics[6]
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Pro | €29/user/mo (14-day trial, 1-user min) | Advanced models (GPT-5, Claude, Gemini, Mistral), custom agents with actions, Slack/GitHub/Drive/Notion connections, Zendesk + Chrome extension, 1GB/user data sources, SOC 2, zero data retention[6] |
| Enterprise | Custom (100+ members) | Everything in Pro plus SSO (Okta, Entra ID, JumpCloud), SCIM, advanced security controls, expanded storage, US/EU hosting, Salesforce tool, priority support[6] |
Licensing model: Per-seat SaaS subscription; platform code is MIT open source but the hosted service is the product[4]
Hidden costs: Programmatic usage (API, Zapier, GSheet) beyond included credits is billed separately at fixed pricing; agent-building time from internal champions is the real adoption cost[6]
Competitive Positioning
Direct Competitors
| Competitor | Differentiation |
|---|---|
| Glean | Glean leads with enterprise search and a horizontal assistant; Dust leads with build-your-own shared agents and per-agent model choice |
| Viktor | Viktor is one Slack-native agent with its own compute environment; Dust is a platform for many purpose-built agents across multiple surfaces |
| Runbear | Runbear focuses on no-code Slack-first assistants; Dust spans more surfaces (web, API, Zendesk, Chrome) with deeper data governance and enterprise compliance |
| OpenAI/Anthropic enterprise tiers | Single-vendor assistants tied to one model family; Dust is model-agnostic with a cross-tool governance layer |
When to Choose Dust Over Alternatives
- Choose Dust when: You want many shared, purpose-built agents across departments, model flexibility, and a governance layer over company data — and have operators willing to build
- Choose Glean when: Enterprise search over existing knowledge is the primary need and agents are secondary
- Choose Viktor when: You want a single capable agent with its own computer executing tasks in Slack, not a platform to build on
- Choose Runbear when: You want the fastest no-code path to a Slack assistant without platform investment
Ideal Customer Profile
Best fit:
- Mid-size to large organizations (Enterprise tier starts at 100 seats) with multiple departments wanting their own agents
- Teams with "AI Operator" types — RevOps, GTM engineers, product marketers — eager to build and iterate on agents
- Companies wanting model optionality rather than betting on one frontier lab
- EU companies valuing a European vendor with EU data hosting
Poor fit:
- Teams wanting a turnkey assistant with zero setup investment
- Organizations requiring self-hosted deployment as a supported path
- Small teams needing enterprise security features (SSO/SCIM) below the 100-seat minimum
Viability Assessment
| Factor | Assessment |
|---|---|
| Financial Health | Strong — $60M+ raised, Sequoia-led Series B, 240% NRR, zero 2025 churn[2] |
| Market Position | Leader in the shared-agent platform niche; 3,000+ orgs, 300K+ agents[1] |
| Innovation Pace | Rapid — model-agnostic platform shipping across six surfaces, active daily open-source commits[4] |
| Community/Ecosystem | Moderate — MIT repo (~1,380 stars) plus strategic ties to Snowflake and Datadog[4] |
| Long-term Outlook | Positive — retention metrics suggest durable expansion, not pilot churn |
Dust's numbers are the story: zero churn and 240% NRR mean customers that adopt the multiplayer model don't just stay, they more than double spend annually.[2] With Sequoia and Abstract leading and Snowflake and Datadog investing strategically, Dust has both the capital and the distribution relationships to consolidate the team-agent category — its main risk is frontier labs bundling "good enough" shared agents into their own enterprise tiers.
Bottom Line
Dust is the most commercially proven independent platform in the team-agent category. Its multiplayer thesis — shared agents built by functional operators, deployed everywhere from Slack to APIs, on whichever model fits — is validated by 3,000+ organizations, 300K+ agents, and retention metrics most SaaS companies would envy.[1]
Recommended for: Organizations with internal champions ready to build department-specific agents on company data, especially those wanting model flexibility and EU-friendly compliance.
Not recommended for: Teams wanting a zero-setup assistant, sub-100-seat companies needing SSO, or organizations requiring supported self-hosting.
Outlook: With $60M+ raised, strategic backing from Snowflake and Datadog, and best-in-class retention, Dust is positioned to define the "multiplayer AI" category it named — provided it stays ahead of frontier labs commoditizing shared agents from above.[3]
Research by Ry Walker Research • methodology