Dust

What it is

Dust is an enterprise platform for building and governing AI agents over your company’s tools and data. You connect sources — Slack, Notion, Google Drive, Salesforce, HubSpot, GitHub, Confluence, Snowflake — then assemble agents in a no-code builder with natural-language instructions, a choice of underlying model (Anthropic, OpenAI, Gemini, Mistral), and a set of tools the agent is allowed to call. The retrieval layer (RAG plus a Table Query mode that runs SQL over structured data) and a dual-layer permission model are what separate it from “point a chatbot at our wiki.” The core is open source under an MIT license; most teams buy the managed cloud. Dust raised a $40M Series B in May 2026 (led by Abstract and Sequoia, with Snowflake Ventures and Datadog), reports 3,000+ organizations and 300,000+ agents deployed, and posted 240% net revenue retention with zero churn in 2025.

Why it shows up across RevOps, Legal, and Recruiting

The same workspace serves all three because the value is one governed agent layer over the SaaS each team already runs:

• RevOps builds an account-research agent that stitches Salesforce notes, Gong calls, and the Slack deal channel into a pre-call brief, or a deal-desk agent that drafts the first pass of an MSA redline.
• Legal ops points an agent at the contract repository and policy wiki to answer “what’s our position on uncapped indemnity?” with the clause and the source, not a hallucinated paraphrase.
• Recruiting runs an agent that summarizes a candidate against the scorecard and the job description, pulling from the ATS and interview notes.

Because agents read and write through the same connector graph, the same permissions, and the same audit trail, a RevOps agent and a Legal agent can be governed under one policy instead of six point tools.

Pricing

• Pro — €29 per user/month (about $31), 14-day trial, no seat minimum. Self-serve, fixed rate for additional model usage, SOC 2, ~1 GB storage per user. This is the transparent entry point and the thing buyers contrast with Glean’s “call us.”
• Enterprise — custom, 100-member minimum. Adds SSO (Okta, Entra ID, JumpCloud), SCIM provisioning, US/EU data residency, advanced controls, and dedicated support. Enterprise model consumption moves to programmatic usage-based pricing, so the all-in number depends on how heavily agents run, not just seat count.

Budget the seat price plus the usage tail. Heavy agentic workflows (long retrieval, frequent tool calls) push token consumption well above what a flat-seat mental model predicts.

Best for

• Ops-led organizations (20–500 seats) that want to build and govern many agents across RevOps, Legal, and Recruiting from one workspace rather than buy a separate AI feature inside each SaaS tool.
• Teams that value model choice, MCP-native tool calling, and EU data residency, and that have someone willing to own agent design and permissions.

Alternatives and when to pick them

• Glean — pick when the primary pain is finding information across fragmented systems, not building agents that take action. Glean’s permission-aware enterprise search and connector breadth lead the category; Dust leads on custom agents and transparent pricing. If 70%+ of the job is “where is the answer,” choose Glean.
• Microsoft 365 Copilot — pick when your team lives in Outlook, Teams, and SharePoint and you want AI inside that suite rather than a separate agent platform. The largest install base by default, and cheaper to adopt when M365 already covers most of the surface area.
• Relevance AI — the fastest-growing no-code “AI workforce” entrant ops teams weigh against Dust. Pick it when the job is multi-agent process automation (an agent team running a back-office workflow end to end) more than a knowledge-grounded assistant over your SaaS.

Watch-outs

• Usage pricing is the real budget, not the seat price. The €29 headline hides model-consumption costs that scale with how much agents retrieve and call tools. Guard: run a 30-day Pro pilot with your actual highest-volume use case before signing Enterprise, and read the token usage off the dashboard rather than estimating from seats.
• Agent quality tracks permission hygiene at the source. An over-shared Drive folder or a stale ACL surfaces to any user who can ask the agent. Guard: audit permissions on the top 5 connected systems before rollout and gate write-enabled tools (anything that edits Salesforce or sends Slack messages) behind an approval step until you trust the agent’s outputs.
• Someone has to own agent design. Self-serve does not mean self-maintaining — ungoverned agent sprawl produces 50 half-working bots and no audit story. Guard: name an owner for the agent library and review new write-capable agents on a fixed cadence instead of letting every team ship its own.