AI Agents & Assistants
Dust
EU-Made Report an errorPanel rating · 6 judges · How to read the stars
Category median
Sovereignty: 1 of 4 dimensions proven
0–5 in half steps. 5 means the rubric's top anchor is met on the evidence.
by Permutation Labs · dust.tt
Report an error on this page Is this your product? →
Link to this page
Add ?theme=dark to the image address for dark backgrounds.
The badge shows the category and the month of the latest evaluation, and EU-Made where
that is the product’s provenance. It never shows a score.
Dust at a glance
- Vendor
- Permutation Labs
- Headquarters
- Not stated on the vendor’s pages as of 26 Oct 2026
- Data hosting
- Default not stated; pages mention the US and the EU source, captured 26 Oct 2026
- Free plan
- Yes, per the vendor’s own page source, captured 26 Oct 2026
- Panel verdict
- The AI panel rates Dust 3.5/5, the mean of 6 judges. Dust, an agent platform from Permutation Labs of Paris, is strongest on integrations and knowledge, where the computed range is 8 to 8: 20+ connectors from Slack and Notion to Salesforce and Zendesk, MCP in both directions, and grounded answers that link back to their sources. how scores are made
- As of
- Vendor pages captured 26 Oct 2026; evaluated by the panel 28 Oct 2026. all sources
Read this page as one judge. Each weighs the same scores by what they care about.
The panel's verdict
Dust, an agent platform from Permutation Labs of Paris, is strongest on integrations and knowledge, where the computed range is 8 to 8: 20+ connectors from Slack and Notion to Salesforce and Zendesk, MCP in both directions, and grounded answers that link back to their sources. Processing and models follows at 7-8 — providers are named, and an EU region at eu.dust.tt can lock the model picker, agent builder and auto models to EU-hosted models, refusing activation while any agent still points at a non-EU model. The lowest ranges are agent capabilities and sovereignty, both 6-7, where lower scores cite human approval described only in general terms, no public information on test-and-evaluation tooling, acknowledged transfers to the United States under standard contractual clauses, and no public information on subprocessor locations. The flagged splits came back empty, and no criterion spans more than a single point; persona-weighted totals run 6.8 to 7.6. Pricing transparency holds at 7: without published per-model credit rates, overage figures or enterprise pricing, annual cost is not computable from public pages.
Speaks for it
- Managed connectors cover Slack, Notion, Confluence, Drive, Gmail, Outlook, HubSpot, Salesforce, Zendesk and more, with data scoping down to folders, sub-pages and specific channels.
- An EU-hosted-models setting restricts the model picker, agent builder and auto models, and refuses to activate while any agent still uses a non-EU model.
- Written commitments that data never trains models, extended by name to every model provider under zero data retention, with SOC 2 Type II published.
- SSO, automated provisioning and deprovisioning, SCIM-synced groups, three roles with additive permissions and audit logs are all documented.
- Public seat prices carry credit allocations — Pro 30€/month or 24€/month billed yearly with 8,000 credits/month, Max 150€/month or 120€/month billed yearly with 40,000 credits/month.
Held against it
- We found no public information on exporting data, agents and configuration when leaving the platform.
- We found no public information on exporting audit logs or streaming them to a SIEM.
- We found no public information on per-action human approval, on test-and-evaluation tooling before release, or on the handling of agent failures.
- We found no public information on per-model credit consumption, the overage cap and rate, or enterprise volume pricing, so annual cost is not computable from public pages.
- The privacy policy acknowledges transfers of personal data to the United States under standard contractual clauses, we found no public information on subprocessor locations, and the EU-hosted-models setting is documented as incompatible with bring-your-own-key.
Best for
- You need agents that search, create and update across company tools like Notion, Confluence, GitHub, Salesforce, HubSpot and Zendesk, with answers that link back to their sources.
- Your workspace must run in the EU with model choice pinned to EU-hosted models and admin caps on which models members may select.
- Your legal review starts from written no-training commitments, zero data retention for named providers, SOC 2 Type II and an enterprise DPA.
- Your IT team needs documented SSO, SCIM provisioning, additive role permissions and audit logs before rollout.
Avoid if
- You require a fully European processing chain end to end — transfers of personal data to the United States under standard contractual clauses are acknowledged, and we found no public information on subprocessor locations or company ownership.
- You want EU-hosted models while supplying your own model provider keys — the two are documented as incompatible.
- You need a computable annual or per-conversation cost for procurement — ask the vendor: the public pages we read do not show it
- You need per-action human approval before agents act in connected systems — the captured pages describe human review only in general terms.
The scores
| Criterion | Panel median |
|---|---|
| Data use, training & retention | 3.5 of 5 |
| Models & processing location | 4 of 5 |
| Admin, identity & access controls | 3.5 of 5 |
| Integrations, connectors & knowledge | 4 of 5 |
| Agent capabilities & guardrails | 3.5 of 5 |
| European sovereignty | 3.5 of 5 |
| Pricing transparency | 3.5 of 5 |
Data use, training & retention
Show reasoningHide reasoning
How this is scored
What happens to prompts, uploaded files, connected data and outputs: whether any of it trains a model (the vendor's or a model provider's), how long it is kept and where logs live, whether zero-retention is available, and whether a data processing agreement covers it.
0 — No statement on training or retention; the privacy policy is a website policy that does not cover the product.
3 — A general statement that business data is "not used for training", with no word on retention, on the model providers behind the product or on logs.
5 — Business data excluded from training by contract, retention periods stated for chats and files, and a DPA offered to business customers.
8 — No-training commitments stated for the vendor and every model provider it routes to, retention configurable by the admin (including deletion of chat history), zero-data-retention terms with the model providers named, and abuse-monitoring retention described.
10 — Data use fully accounted for: training, retention, logging and abuse monitoring stated per model provider, customer-controlled retention and deletion with an audit trail, outputs and connected data treated as customer data in the DPA, and an independent audit (SOC 2, ISO 27001 or ISO 42001) covering the product.
The Head of IT
The security page commits that our data is never used to train models, and the privacy policy states every foundational model provider (OpenAI, Anthropic, Mistral, Google, Fireworks) is prohibited from training on customer data and applies zero data retention, with a DPA on enterprise terms and SOC 2 Type II claimed. I found no public information on retention periods for chats and files on the platform, admin deletion of chat history, or abuse-monitoring retention. 3 4 2 1
The Data Protection Officer
The vendor commits that data is never used to train models and extends that prohibition by name to every foundation model provider behind the product, backed by a zero-data-retention policy for those providers — the exact chain of commitments I need before signing. Retention policies are customizable by the customer, a DPA is offered on enterprise plans and SOC 2 Type II is published; we found no public information on abuse-monitoring retention or on admin-side deletion of chat history. 1 2 3 4
The Automation Lead
The security pages commit that our data is never used to train models, name every foundation model provider behind the product as contractually prohibited from training, and state a zero-data-retention policy under which those providers do not log for human review or save content to disk, with admin-configurable retention policies and SOC 2 Type II certification on top. Retention periods are stated for contract and marketing data rather than chats and connected files, and abuse-monitoring retention and deletion with an audit trail are not described, which caps it below the top. 1 2 3 4
The Customer-Service Director
The security page states your data is never used to train models and no data is stored by third-party model providers, and the privacy policy extends the prohibition to the named providers OpenAI, Anthropic, Mistral, Google and Fireworks under zero-data-retention terms, with custom data retention policies for admins, a DPA on the Enterprise plan and SOC 2 Type II. That clears the bar for our compliance review to begin; we found no public information on abuse-monitoring retention or explicit deletion of chat history. 3 4 1 2
The Works Council Member
Training is addressed in writing: our data never trains the vendor's models, the named model providers are prohibited from using customer and personal data for training, and a zero data retention policy is stated for them, with custom retention policies and SOC 2 Type II on top. We found no public information on abuse-monitoring retention or on admin-driven deletion of chat histories, and the DPA with custom legal terms is reserved for the enterprise tier — points we would need settled in the agreement before staff prompts flow. 1 2 3 4
The Skeptic
The security page states "Your data is never used to train models" and "No data is stored by third-party model providers", and the privacy policy goes further: the model providers (OpenAI, Anthropic, Mistral, Google, Fireworks) are prohibited from training on customer data and apply a stated Zero Data Retention policy. Retention is admin-configurable, a DPA is offered (enterprise only) and SOC 2 Type II is claimed — but we found no public information on abuse-monitoring retention or on deletion of chat history. 1 2 3 4
Models & processing location
Show reasoningHide reasoning
How this is scored
Which models sit behind the product and who runs them, where inference and storage happen, whether the customer can pin processing to the EU, choose or exclude model providers, bring its own model or key, or run the product in its own cloud or on premises.
0 — The models behind the product and the processing location are not stated.
3 — The model family is named, but where inference runs and through which provider (the model maker, a hyperscaler, the vendor) is unstated.
5 — Models and their providers named, with an EU processing option stated for storage and at least the default model.
8 — Each available model listed with provider and inference region, EU-only processing selectable by the admin and enforced across models, model providers excludable, and bring-your-own-key or own-cloud deployment documented.
10 — Processing fully under the customer's control and documented: EU-only or on-premises inference for every feature including agents and connectors, open-weight or self-hosted models supported, data flows per feature published, and a residency commitment in the contract rather than on a marketing page.
The Head of IT
Models and providers are named, and the EU region has a documented admin setting that restricts the model picker and Agent Builder to EU-hosted models, auto models included, refusing activation if an agent still points at a non-EU model. Admins cap models per member and manage workspace model providers, bring-your-own-key is acknowledged and single-tenant deployment exists at enterprise level, though the EU setting is incompatible with bring-your-own-key and requires Dust's model hosting on a credit plan. 5 6 3 2 8
The Data Protection Officer
Each model is listed with its provider and an EU flag where it runs EU-hosted, and an admin can pin the workspace to EU-hosted models only — a switch that refuses to enable while any active agent still uses a non-EU model, so it reaches agents, not just the chat window. Providers are selectable, bring-your-own-key and single-tenant enterprise deployment exist, but the EU-only setting is not compatible with own provider keys, and we found no public information on where connectors and tools process data. 2 3 5 6
The Automation Lead
Every provider's models are listed and split into EU-hosted and non-EU-hosted lists, and an EU-hosted-models-only setting in the admin panel restricts the model picker, the Agent Builder and the auto models for a workspace on the EU region — this is enforced processing choice, not a marketing line. The EU path requires a credit-based plan on the vendor's own model hosting rather than our provider keys, and we found no public information on on-premises deployment or a residency commitment at contract level, so it sits just short of full customer control. 2 3 4 5 6
The Customer-Service Director
Models are named per provider and flagged EU-hosted or not in the picker, an EU workspace at eu.dust.tt can be locked to EU-hosted models only from the admin panel, and admins can cap which models and reasoning efforts members select. One caveat for my EU contacts: the EU-hosted-models setting requires Dust's own model hosting and is not available when using your own model provider keys. 5 6 2 3 4
The Works Council Member
Model families and providers are public, EU-hosted variants are listed per provider and flagged in the picker, and an EU workspace can lock the model picker, the agent builder and the auto models to EU-hosted models, with admins able to cap which models each member selects. The EU lock only applies on a credit-based plan using the vendor's own model hosting rather than members' own provider keys, and single-tenant deployment is an enterprise feature. 2 3 5 6
The Skeptic
Every model provider is named (OpenAI, Anthropic, Google, Mistral and more, 20+ models), and an EU region can restrict agents to models Dust hosts in the European Union — with a check that refuses the setting while any agent still uses a non-EU-hosted model. We found no public information on where the non-EU models actually run, bring-your-own-key is explicitly incompatible with the EU-hosted-models setting, and single-tenant deployment is enterprise-only. 2 3 5 6
Admin, identity & access controls
Show reasoningHide reasoning
How this is scored
What the IT department controls: SSO and SCIM provisioning, roles and groups, which models, agents and connectors each group may use, retrieval that respects the permissions of the source system, audit logs, usage reporting, and policies such as data-loss prevention — and whether reporting on staff use is scoped and documented.
0 — No admin features documented; accounts are individual sign-ups.
3 — A team workspace with an owner and member invitations, but no SSO, no roles and no audit log.
5 — SSO, admin and member roles, user management, and an admin view of usage.
8 — SAML/OIDC SSO with SCIM provisioning, group-based permissions for models, agents and connectors, retrieval that inherits source-system permissions, exportable audit logs, and usage analytics with documented scope.
10 — Governance a regulated company can operate: fine-grained policies (DLP, content filters, prompt and output logging rules), audit logs streamed to a SIEM, per-agent ownership and approval, usage reporting that can be aggregated or pseudonymised for works-council agreements, and every admin action logged.
The Head of IT
This is close to rollout-grade: SSO, automated provisioning with SCIM-synced groups, additive group permissions with Everyone/Groups/Admin-only modes, a dual-layer permission model with admin-gated overrides and zero privilege escalation, audit logs and governance analytics. I found no public information on exporting audit logs or streaming them to a SIEM, and retrieval is governed by admin-scoped connections and space membership rather than a stated inheritance of the end user's permissions in the source system; model capping is documented per member, not per group. 1 8 7 6 3
The Data Protection Officer
SSO with automated provisioning, SCIM-synced groups, three workspace roles with additive permissions over agents, skills, spaces and tools, admin caps on which models members may select, and audit logs cover most of what an IT department needs to operate. We found no public information on exporting audit logs, on retrieval that inherits the permissions of the connected source system, or on data-loss-prevention policies. 1 6 7 8
The Automation Lead
SSO with SCIM-provisioned groups, three roles plus additive granular permissions covering who may create agents, manage tools and spend workspace credits, per-space restricted access to tools, admin-level configuration of third-party credentials, an audit-log toggle and role-scoped analytics — this is the governance model I would hand to IT before an agent touches the CRM. We found no public information on exporting or streaming audit logs to a SIEM or on data-loss-prevention policies. 1 7 8
The Customer-Service Director
SSO, automated provisioning and deprovisioning, SCIM-synced groups, three workspace roles with granular additive permissions, a dual-layer permission model separating what agents can access from who can use them, audit logs and usage analytics including costs are all documented in depth. We found no public information on retrieval inheriting permissions directly from the source systems, or on exporting audit logs to our SIEM. 1 7 8 6
The Works Council Member
SSO, automated provisioning and deprovisioning, SCIM-synced groups, three roles with additive granular permissions, group-level control over agents, tools and connectors, audit logs, and analytics on activity, usage, costs and ROI are all documented — solid IT governance. For our co-determination file, though, we found no public information on usage reporting that can be aggregated or pseudonymised for a works agreement, and audit logs can be enabled or disabled by an admin, so prompt-logging rules would have to be fixed in writing contractually. 1 6 7 8
The Skeptic
SSO, automated provisioning and deprovisioning, SCIM-synced groups, three roles with additive granular permissions, admin caps on which models and reasoning efforts members may select, audit logs and usage analytics are all documented. We found no public information on the SSO protocol, on exporting audit logs, or on retrieval that inherits the permissions of the source system. 1 6 7 8
Integrations, connectors & knowledge
Show reasoningHide reasoning
How this is scored
How the product reaches company knowledge and systems: connectors to file stores, wikis, mail, CRM, ticketing and ERP, how fast and how completely they sync, the API, support for open protocols such as MCP, and how the assistant cites the sources it answers from.
0 — No integrations; users paste text or upload files one at a time.
3 — File upload and one or two connectors (for example a drive), with no API and no statement on sync or citations.
5 — A connector library covering the common file stores, wikis and collaboration tools, answers citing sources, and a documented API.
8 — Connectors to file stores, wikis, mail, CRM and ticketing with incremental sync, custom connectors or MCP support, an API and webhooks for embedding assistants and agents in other tools, and source citations on every grounded answer.
10 — The product sits in the company's systems: read and write connectors with documented scopes, MCP servers and clients, contact-centre or telephony integration where relevant, a knowledge layer that can be audited and corrected, and the data, agents and configuration exportable when the customer leaves.
The Head of IT
Connectors cover file stores, wikis, mail, CRM and ticketing, with MCP both as client (remote MCP servers as tools) and as a server, a developer API with conversation and data-source APIs on enterprise, webhooks from GitHub, Jira, Linear and Zendesk, and automation platforms. Grounded answers link back to sources, and knowledge scope goes down to folders, sub-pages and specific channels with label filtering; I found no statement on incremental sync. 2 11 10 12 7
The Data Protection Officer
Fully managed connections sync Slack, Notion, Drive and more with folder- or document-level scoping, a developer API plus conversation and data-source APIs and webhooks let agents live inside other tools, and MCP works both ways — remote servers as tools and a Dust MCP server for workspace management. Grounded answers carry links to their sources and agents read and write in GitHub, Confluence, Salesforce and Zendesk; we found no public information on incremental sync intervals or on exporting data and agents when leaving. 2 7 10 11 12
The Automation Lead
Managed connections to Slack, Notion, Google Drive and more with folder- and document-level scoping, read-and-write tools into Confluence, Notion, GitHub, Gmail, Outlook, Slack, Salesforce, HubSpot, Zendesk and Calendar under personal or workspace credentials, remote MCP servers attachable as tools plus MCP server tools of its own, a conversation and data-source API, and webhooks from GitHub, Jira, Linear, Zendesk or any custom HTTP source. Grounded answers come back with links to the sources, and label filtering and folder navigation make the knowledge layer refinable. We found no public information on exporting data, agents and configuration when leaving the platform. 2 7 9 10 11 12
The Customer-Service Director
Connectors cover file stores, wikis, mail, CRM (Salesforce, HubSpot) and ticketing with Zendesk ticket search, metrics and reply drafting, plus any tool via MCP, a developer and conversation API, webhooks, and source links on grounded answers. For my contact centre I found no public information on telephony or contact-centre platform integration, so the voice channel would sit outside this. 2 7 10 11 12
The Works Council Member
Twenty-plus connectors including Slack, Notion, GitHub and Drive, any tool via MCP, data scoping at folder or document level down to sub-pages and specific channels, developer, conversation and data source APIs, and agent tools that search, update and create in Confluence, Notion, GitHub, Salesforce, HubSpot and Zendesk. Grounded answers link back to their sources, which we need for verification. We found no public information on sync frequency or on exporting data, agents and configuration when a customer leaves. 2 7 10 11
The Skeptic
The connector library reaches file stores, wikis, mail, CRM and ticketing — Slack, Notion, Confluence, Drive, Gmail, Outlook, HubSpot, Salesforce, Zendesk — with tools that search, create and update in those systems, MCP in both directions (remote MCP servers as tools, plus a Dust MCP server), a developer API, and grounded answers that link back to their sources. We found no public information on incremental sync behaviour or on exporting data, agents and configuration when the customer leaves. 2 7 8 10 11
Agent capabilities & guardrails
Show reasoningHide reasoning
How this is scored
What an agent can actually do and where it stops: multi-step tasks, tool use and actions in other systems, scheduling and triggers, human approval before consequential actions, run logs, evaluation and testing before release, and the limits the vendor states rather than hides.
0 — No agents; the product is a chat window, whatever the marketing says.
3 — Saved prompts or custom assistants with instructions and files, but no actions in other systems and no run history.
5 — Agents that call tools or APIs and run multi-step tasks, built without code, with a log of each run.
8 — Agents that act in connected systems with scoped permissions, human approval configurable per action, triggers and schedules, test and evaluation tooling before release, and documented limits and failure handling.
10 — Agents a business can put in production and account for: versioning and staged rollout, evaluation against test sets, monitoring of quality and cost per agent, handover to a human with full context, every action attributable to the agent and the person who deployed it, and the vendor's own statement of what agents must not be trusted with.
The Head of IT
Agents act in connected systems on Personal or Workspace credentials configured at admin level, triggers run on schedules and webhooks with each run recorded as an inspectable conversation, and ownership and credit consumption are attributable per run with tools restrictable per space. I found no public information on configurable human approval before consequential actions or on test and evaluation tooling before release, and documented limits are partial (a webhook rate limit of 42 runs per 24 hours, EU-model eligibility checks). 11 12 1 5
The Data Protection Officer
Agents act in connected systems under tool permissions configured at admin level with per-space restrictions, run on schedules and webhooks with a stated default rate limit, and every run lands as a conversation the owner can open and review. The captured pages mention human review only in general terms, and we found no public information on per-action approval configuration or on test and evaluation tooling before release. 1 7 11 12
The Automation Lead
Agents act in connected systems with tools restricted per space and credentials configured at admin level, run on plain-language schedules and webhooks with a documented default rate limit of 42 runs per 24 hours, and every run lands as an openable conversation with its credit consumption reviewable — run logs and cost attribution exist. We found no public information on configurable human approval before a consequential action, on test or evaluation tooling before release, or on versioning and staged rollout, which is exactly what I would need before company-wide deployment. 1 2 11 12
The Customer-Service Director
Agents act in connected systems under personal or workspace credentials configured at admin level, run on schedules and webhooks — the documented example even summarizes yesterday's new support tickets every weekday — and each run leaves a conversation you can open to see what it did. We found no public information on test and evaluation tooling before release, on human approval configured per action, or on handover to a human with full context, which is exactly what I need before letting an agent loose on forty seats. 11 12 1 2
The Works Council Member
Agents run multi-step, act in connected systems on personal or workspace credentials that must be configured at admin level, and run on schedules and webhooks, with each run recorded as a reviewable conversation, credit consumption inspectable, and webhook sources restricted to admin creation. Human approval appears only as a general "human review" phrase rather than a per-action approval setting, and we found no public information on test or evaluation tooling before release — both are conditions we would set before agents act on colleagues. 1 11 12
The Skeptic
Agents act in connected systems through admin-configured tools with per-space restrictions and personal or workspace credentials, run on schedules and webhooks, and each run creates a conversation you can open to inspect, with per-trigger credit consumption reviewable. We found no public information on human approval before consequential actions, on test-and-evaluation tooling before release, or on how failures are handled. 1 2 11 12
European sovereignty
panel opinion
Show reasoningHide reasoning
How this is scored
Who the contracting entity and the parent company are, where customer data is hosted, and who the subprocessors are — model providers above all, then hosting, search, telephony and email. Independently sourced by the sovereignty pipeline; for an AI product the model provider is a subprocessor, so a European interface over a non-European model counts as such.
0 — Non-EU vendor and contracting entity, hosting unstated, subprocessors (including model providers) unnamed.
3 — A privacy policy exists, but hosting location is unstated or non-EU and the model providers behind the product are not named as subprocessors.
5 — EU hosting stated and an EU contracting entity, but prompts reach non-EU model providers or other subprocessors without a stated safeguard or EU region.
8 — EU hosting on named infrastructure, EU contracting entity, every subprocessor listed with its location including model providers, EU inference available, and a DPA offered.
10 — Sovereign end to end and evidenced: European ownership, entity, hosting and model processing, every subprocessor European or optional, data export and deletion documented, and a security or certification statement published rather than implied.
The Head of IT
The contracting entity is Permutation Labs in Paris, an EU region hosts workspaces with EU-hosted model processing, and the privacy policy names its subprocessors — OpenAI, Anthropic, Mistral, Google, Fireworks for models, Google Cloud Platform and QDrant Cloud for infrastructure — with standard contractual clauses covering US transfers and a DPA on enterprise plans. Subprocessor locations are not listed individually and ownership is not published, so I cannot confirm sovereign end to end. 4 5 2 3
The Data Protection Officer
The contracting entity is Permutation Labs at an address in Paris, workspaces can be hosted in the EU with EU-hosted inference at eu.dust.tt, the model providers are named, a DPA is offered and SOC 2 Type II is published. Prompts still reach US-headquartered providers (OpenAI, Anthropic, Google, Fireworks) unless an admin actively restricts models, ownership is not documented, and the hosting subprocessors Google Cloud Platform and QDrant Cloud are named without their regions. 2 3 4 5
The Automation Lead
A French contracting entity in Paris, a documented EU region with EU-hosted model lists, model providers named as subprocessors under a zero-data-retention commitment, and a DPA offered on enterprise terms. Personal data may still be transferred to the United States with standard contractual clauses as the stated safeguard, and the hosting infrastructure named (Google Cloud Platform, QDrant Cloud) does not come with per-region locations, so it stops short of end-to-end European processing. 2 3 4 5
The Customer-Service Director
The contracting entity is Permutation Labs at an address in Paris, workspaces can be hosted in an EU region restricted to EU-hosted models, model providers are named as subprocessors, and a DPA is offered. The privacy policy acknowledges transfers to the United States under adequacy decisions or standard contractual clauses and the provider list is dominated by US-headquartered firms, so the EU lock is optional configuration rather than the default. 4 5 2 3
The Works Council Member
The contracting party is Permutation Labs of Paris, France, trading as Dust, an EU region with EU-hosted models is available, a DPA is offered and SOC 2 Type II is stated. The model providers are named — OpenAI, Anthropic, Mistral, Google, Fireworks — with several of them US-headquartered, hosting runs on Google Cloud Platform and QDrant Cloud, and transfers to the United States are acknowledged with standard contractual clauses as safeguard; ownership is not publicly stated, so a fully European chain end to end is not demonstrated on these pages. 2 4 5
The Skeptic
The contracting entity is Permutation Labs at 86 avenue de Wagram, Paris, France, hosting is offered in the EU and the US, an EU region with EU-hosted models exists, and the subprocessors — every model provider plus Google Cloud Platform and QDrant Cloud — are named. Subprocessor locations are not stated, the privacy policy itself records transfers of personal data to the United States under SCCs, and the EU-only mode requires a credit-based plan using Dust's model hosting rather than your own provider keys. 2 3 4 5
Pricing transparency
Show reasoningHide reasoning
How this is scored
Whether a company can compute the real annual cost from public pages alone: price per seat, per agent, per conversation or per usage unit, what is included in each tier, usage limits and overage, which models, connectors or agents cost extra, minimum seats and term.
0 — No public prices at all; every customer is a sales call.
3 — A starting price or "from" figure exists, but usage limits, included models and what scales the cost are unclear.
5 — Prices per tier are public with the main inclusions, but usage limits, overage, agent runs or enterprise features (SSO, audit logs, EU processing) are unpriced or "on request".
8 — Every tier priced publicly with seat or usage units, usage limits and overage rates, add-ons listed, minimum seats and term stated.
10 — Complete price computability: annual cost derivable for a given number of seats, agents or conversations, with every usage rate, model surcharge, add-on and support tier and the cancellation and data-return terms published.
The Head of IT
Per-seat prices are public and specific — Max at 120€/seat/mo billed yearly with 40,000 credits, Pro at 24€ with 8,000, a 20% annual saving, no rollover, capped admin-enabled overage, mid-cycle proration and seat-reassignment terms. For an eight-hundred-seat rollout the published plans stop at teams of 100 people, and enterprise volume pricing, overage rates and the model access tiers some models require are unpriced, so the real annual cost is not computable from public pages alone. 2 1 5
The Data Protection Officer
Public per-seat prices cover every self-serve tier with monthly and yearly figures, monthly credit allocations, reset rules, capped admin-enabled overage, proration and seat reassignment, and VAT excluded is stated plainly. A buyer still cannot compute the full annual cost: enterprise volume pricing and the overage cap are not quantified, and model access tiers are referenced without prices. 2 5
The Automation Lead
Free, Pro and Max seat prices are published with their credit allocations, the annual discount, monthly reset without rollover, mid-cycle proration, seat reassignment rules and an overage policy that is capped and admin-enabled, all excluding VAT, and trigger runs are priced as ordinary agent runs with a configurable payer. The real annual usage cost is not computable from public pages: we found no public information on how many credits each model consumes, on the overage cap or rate, or on enterprise and single-tenant pricing beyond volume pricing. 2 12
The Customer-Service Director
Seat prices and credit allowances are public per tier with monthly and yearly figures, and rollover, overage behaviour, mid-cycle proration, seat reassignment and VAT exclusion are all stated plainly. What the CFO cannot yet compute is a cost per conversation: credit consumption per message varies by model and actions without a published per-model rate, the overage cap carries no public rate, and the Enterprise tier is described as shared credits and volume pricing without figures. 2 12 1
The Works Council Member
Pro is 30€/month or 24€/month billed yearly with 8,000 credits, Max is 150€/month or 120€/month billed yearly with 40,000 credits, and the free tier, 20% annual saving, VAT exclusion, no credit rollover, mid-cycle proration and admin-enabled capped overage are all public. The overage rate, the pricing of model access tiers and enterprise volume pricing are not published figures, so the true annual cost cannot be computed from the public pages alone. 2 5 12
The Skeptic
Every paid tier carries a public seat price with its credit allocation — Pro 30€/month or 24€/month billed yearly with 8,000 credits/month, Max 150€/month or 120€/month billed yearly with 40,000 credits/month — plus rollover rules, mid-cycle proration, overage mechanics and VAT treatment. We found no public information on the overage cap amount, the credit cost of each model, or enterprise pricing beyond "volume pricing". 1 2 5
European sovereignty — proven facts
1 of 4 dimensions provenBuilt only from facts shown on the vendor's own pages. A dimension we could not prove is left open, not scored as zero.
| Legal entity | Not determined ⚠ unverified | — | uncited Report an error |
|---|---|---|---|
| Ownership | Not determined | — | uncited Report an error |
| Data residency | Not determined ⚠ unverified | — | uncited Report an error |
| Subprocessors | US CLOUD Act reach ⚠ unverified | 0/2 pts | 4 Report an error |
Questions about Dust
Is Dust European?
Yes. whats-best.ai classifies Dust as EU-Made.
Where does Dust store data?
The default storage location for customer data is not stated on the vendor’s pages as of 26 Oct 2026. The vendor’s pages mention the US and the EU as hosting locations, per the vendor’s page captured 26 Oct 2026.
Does Dust have a free plan?
Yes. Dust has a free plan, per the vendor’s page captured 26 Oct 2026.
What are the best alternatives to Dust?
As of 28 Oct 2026, the panel rates Langdock (4/5), Unique AI (3.5/5) and Omnifact (3.5/5) highest among the other 15 AI Agents & Assistants products evaluated. Ratings that look equal are rounded to half stars; the order uses the unrounded rating, then the name. The highest-rated European alternative is Langdock (EU-Made). Ratings are the AI panel’s opinion, formed from public evidence: how scores are made. The full list, with European products first: Dust alternatives.
Where this could be wrong
- Evidence ages. The oldest capture behind this page is from 26 Oct 2026. Vendors change pricing and policies without notice; every fact reflects its source as of the capture date shown in the registry.
- Weak sourcing — Legal entity, Data residency. Not confirmed on the vendor’s own pages as captured.
- Weak sourcing — Subprocessors. Mistral is EU-based, but OpenAI, Anthropic, Google and Fireworks are US-headquartered, and the same privacy policy's security section also names Google Cloud Platform and QDrant Cloud as hosting infrastructure.
- AI can misread a source. Extraction and judgement are automated; a citation guarantees traceability, not infallibility. If something here is wrong, say so — no account needed, every report is decided within 5 business days, and accepted corrections are published.
What we left out
A claim that does not survive our checks costs us the claim, not the page. This is what was taken off this one.
- 11 integrations facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 5 product facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 4 compliance facts could not be confirmed on the vendor’s page as captured and were left out of this page and of the panel’s material. Know more? Tell us
- 1 legal fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 1 pricing fact could not be confirmed on the vendor’s page as captured and was left out of this page and of the panel’s material. Know more? Tell us
- 2 sovereignty dimensions could not be confirmed on the vendor’s own pages and are shown as unknown. Know more? Tell us
Sources (12)
The pages every claim on this page was read from — each one checked, dated, and kept verifiable.
- 1 Vendor homepage dust.tt Checked 26 Oct 2026 Details →
- 2 Pricing page dust.tt Checked 26 Oct 2026 Details →
- 3 Security page dust.tt Checked 26 Oct 2026 Details →
- 4 Privacy policy naming the legal entity dust.tt Checked 26 Oct 2026 Details →
- 5 Models & processing location — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 6 Models & processing location — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 7 Admin, identity & access controls — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 8 Admin, identity & access controls — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 9 Integrations, connectors & knowledge — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 10 Integrations, connectors & knowledge — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 11 Agent capabilities & guardrails — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →
- 12 Agent capabilities & guardrails — found from sitemap docs.dust.tt Checked 26 Oct 2026 Details →