Dust

Dust is an enterprise multi-agent collaboration platform built by the Paris-based team Permutation Labs, centred on "Multiplayer AI": it puts people, Agents, company knowledge and tools into one governable shared workspace. It targets mid-size and large enterprises, letting Agents be grounded in company data, shareable and auditable, rather than acting as a single-point chat assistant. It is unrelated to the game, novel, messaging app and Web3 projects that share its name.

Core capabilities

• No-code Agent building: natural language + data sources + tools, published in minutes and shared across the team
• Multi-agent orchestration: schedule/event/Webhook triggers, with Agent outputs chained together
• Model neutrality + MCP: connects to 20+ frontier and open-source models, and tools can be integrated through MCP servers
• Collaboration primitives: Pods shared spaces, interactive Frames output, self-improving Skills/Memory
• Enterprise governance: permissions, audit logs, cost analytics, and replayable run inspection for debugging
• 70+ connectors: Slack, Notion, Google Drive, Salesforce and more (figures vary)

Where it fits Suited to Ops/Support/Sales/Marketing/Data teams building a grounded, shareable and governable Agent layer over internal knowledge; mid-size and large companies with roughly 5-500+ seats fit best. Individual users, or small teams that only need flat-rate billing, will find it heavy and expensive.

Differentiators Compared with Glean's company-wide search or Copilot's Microsoft stack, Dust's moat is multiplayer (shared human-Agent context) + model neutrality + MCP + European data residency and governance. It is not enterprise search - it has fewer connectors and replaces source-system ACLs with its own Space permission model.

Editorial take A distinctive position with strong compliance backing (Sequoia-led funding, SOC 2 Type II, GDPR), suited to a cross-departmental Agent layer in mid-size and large enterprises. Buyers should verify for themselves: dual credit + per-seat billing makes TCO hard to predict, the self-service tier has many limits, and the learning curve is steep; review samples are thin and the shared brand name adds noise, so we draw no conclusion about "concentrated praise or criticism".