Dust: a governed RFQ copilot, not an autonomous quoting authority

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

An industrial RFQ can require drawings, revisions, material notes, capability records and commercial context. The difficulty is finding permitted evidence, distinguishing a current version from a similar old job and routing unresolved decisions to the person who owns them.

What Dust does

Dust, founded in Paris in early 2023 and now operating across Paris and San Francisco, began with multi-model enterprise assistants connected to company data. By June 2024 it supported Notion, Google Drive, Intercom and Slack, an application programming interface for custom connectors and use through its web interface or Slack. In 2026, the company describes “multiplayer AI”: shared workspaces where people and agents use common context, tools and prior work. 1 5 12

The public product advertises 20-plus models, 70-plus connections, Model Context Protocol (MCP), scheduled or event-driven workflows and enterprise controls. Dust says more than 3,000 organisations have deployed more than 300,000 agents, naming Clay, Profound, Persona and Doctolib. These are company-reported adoption figures, not an independently audited productivity study. 1 2

Retrieval has a permission boundary

Administrators can scope connected content at folder or document level, place it in open or restricted Spaces and define membership. Users can create or interact with agents only within Spaces they can access. For an API data source, Dust chunks text, creates embeddings, indexes vectors and metadata, and supports document-level replacement by document_id; queries are embedded and matched to relevant chunks. 9 10

That makes Dust a plausible knowledge and RFP copilot: it could find prior drawings, capability evidence, approved templates, customer context and missing fields, then draft a response or route an exception. It does not establish that Dust understands tolerance stacks, validates material compatibility, calculates a binding price, reserves capacity or issues an approved offer. Enterprise pages claim read/write integrations, REST APIs, webhooks and OAuth2, but write capability is not commercial authority. 3 11

Reported outcomes need denominators

TechCrunch reported that 75% of Qonto’s 1,600-person team used assistants monthly in 2024; it also named Alan, Pennylane, Watershed and PayFit. These are customer or company disclosures, not causal ROI or accuracy measures. 5

A 2026 TNW account reports company-supplied figures of 41,000 monthly active users, 70% weekly active usage and zero customer churn in 2025. It says Vanta’s 46-person revenue team estimated more than 400 hours saved per week; Watershed reduced a recurring data-mapping task from two to three hours to a few minutes at 78% success; and Qonto estimated about 50,000 hours saved per year across more than 50 specialised agents and 1,000-plus daily users. Baselines, task samples, error definitions and independent verification are not provided. 6

Dust’s sales material claims 90% faster RFP response, eight hours saved weekly per sales representative and 43% of Kyriba employees saving more than three hours weekly on RFPs. Its analytics material claims a 90% reduction in Wakam’s partner-intelligence processing. These are provider/customer claims and useful pilot hypotheses, not general benchmarks: methods, populations, error rates and realised-versus-estimated savings are not disclosed. 7 8

An illustrative manufacturing pilot

For an illustrative workshop, start with one governed RFQ workflow. Connect only approved customer, enterprise resource planning, product-lifecycle and document records. Preserve part identities, drawing revisions, units, tolerances, price validity and capacity status. Ask Dust to retrieve evidence, draft an assumption-tagged response, identify missing fields and route exceptions to engineering or sales. The commercial owner remains authoritative for price, material, compliance, capacity and any binding offer.

Replay RFQs containing stale revisions, duplicate documents, mixed units, incompatible materials, prompt-injection content and retry conditions. Measure authoritative-record retrieval, wrong-revision rate, first-pass accuracy, escalation rate, review minutes, credits consumed and accepted outputs. Use restricted Spaces, least-privilege connections, source citations, write allow-lists, idempotency keys and human approval. Dust’s SOC 2 Type II, GDPR, residency and governance claims reduce exposure, but do not remove dirty-data or authority risks. 1 2 3 9 10

Cost and decision

Public Business pricing is Free at €0 with 500 lifetime credits, Pro at €30 monthly or €24 monthly annually with 8,000 credits, and Max at €150 monthly or €120 monthly annually with 40,000 credits. Programmatic usage is $0.01 per credit; Enterprise pricing is custom. Add data cleanup, connector work, permission design, evaluation, monitoring, migration and human review to total cost. 3 4

Dust is credible for governed retrieval, shared workflows and tool-connected drafts. The evidence supports a labelled RFQ pilot, not autonomous quoting. No reviewed source verifies industrial RFQ accuracy, binding pricing, capacity reservation or independent ROI. 1 2 3 6 7 10

Sources and next reading

• Dust web app and product overview — public overview; an account is required for workspace use.
• Dust pricing and self-serve plans — public pricing; Business is self-serve and Enterprise is sales-led.
• Dust sales-agent workflows — public demos; live use requires an account and data connections.
• Dust developer data-source documentation — public API documentation; execution requires an API key and workspace.
• Dust enterprise deployment overview — public overview; deployment and custom terms require vendor engagement.

Explore OfferSignal · Focused Business & AI Workshop · All research

References