Dust Review 2026: The No-Code Agent Platform for Teams
Reviewed by JustPrompt Editorial Team · Updated July 29, 2026
4.2/5
We tested Dust's agent builder, integrations, and EU-first governance stack to see which teams benefit — and how it compares to Copilot and Cohere.
- No-code agent builder, no developers required
- 61+ integrations across real company tools
- Strong EU sovereignty and compliance stack
- Multi-model support avoids single-vendor lock-in
- Pricing structure has shifted, confusing to verify
- Real onboarding friction on large document sets
- Value depends entirely on data source quality
Overview
Dust is the platform betting that the real bottleneck in enterprise AI isn't model intelligence — it's the fact that most companies' actual knowledge is scattered across Slack threads, Notion pages, Drive folders, and GitHub repos that no chatbot can see. Founded in Paris and headquartered there still, Dust builds a no-code agent platform that connects directly to a company's real tools — 61-plus integrations spanning Slack, Google Drive, Notion, GitHub, and more — and lets teams build custom AI agents grounded in that actual internal knowledge rather than general web training data. By its May 2026 Series B announcement, Dust reported over 3,000 organizations, roughly 41,000 monthly active users, more than 300,000 deployed agents, and — a genuinely notable retention signal in enterprise software — zero customer churn through 2025.
The pitch sits in the same territory this catalog has already explored from two different angles: Cohere's purpose-built retrieval architecture for enterprises, and Mistral's European-jurisdiction sovereignty case. Dust combines threads from both. Like Cohere, its whole design centers on grounding AI answers in an organization's real documents rather than treating retrieval as an afterthought — but where Cohere sells the underlying API components for engineering teams to assemble, Dust sells the assembled, no-code product directly to the business teams who'll actually build and use the agents, no developer required. And like Mistral, Dust's Paris headquarters and EU-first architecture make it one of the leading non-Microsoft options for European organizations that want AI agents without American jurisdiction attached — reviewers have specifically flagged it as a genuine alternative to Microsoft Copilot's enterprise tier (reviewed elsewhere in this catalog) for exactly that reason.
What you actually get is a workspace where anyone — not just engineers — can point an agent at specific data sources (this Slack channel, that Notion space, these GitHub repos), define its behavior and tone, and deploy it for a specific job: a company-knowledge Q&A assistant for new hires, an engineering-documentation search bot, a customer-support agent grounded in actual product docs rather than a generic script. Crucially, Dust is deliberately model-agnostic — agents can run on GPT, Claude, or Gemini models depending on the task, the same multi-model pragmatism this catalog has praised at Notion and several coding tools, letting an organization hedge against any single provider rather than marrying one lab's roadmap.
The concrete use cases reviewers keep returning to sketch the product's actual shape better than any feature list: a company-knowledge Q&A assistant that lets new hires ask "how does our expense policy work" and get an answer sourced from the actual current handbook rather than whatever a person half-remembers; engineering documentation search that spans scattered repos and wikis so a developer isn't hunting through six different tools to find how a service was configured two years ago; and onboarding assistants that answer the dozens of small procedural questions a new employee would otherwise interrupt a colleague to ask. None of these are glamorous AI demos — they're the unglamorous, compounding friction every growing organization accumulates, and Dust's bet is that solving it well, repeatedly, across a whole company's internal knowledge is worth more than any single flashy capability.
The governance layer is where Dust makes its case most seriously to the buyers who actually sign enterprise contracts: SSO, granular per-data-source permissions, full audit logs, configurable retention controls, EU data residency, ISO 27001 and SOC 2 certification, and — a detail worth underlining given how many AI vendors bury this — a no-training agreement with the underlying model providers, meaning the company data Dust's agents read doesn't feed back into anyone's training pipeline. For knowledge-intensive teams in regulated or IP-sensitive industries, that combination of real integration depth and real compliance infrastructure is a coherent, credible pitch, not a checkbox exercise.
The honest complications, per multiple independent reviews converging on the same points: onboarding has real friction. Initial indexing of large documentation sets can take days rather than minutes, a genuine "learning curve" is commonly cited by new users unfamiliar with agent-building workflows, and per-seat pricing that starts reasonable can scale quickly as a team grows or as usage-based credit limits get hit — a pattern this catalog has now seen at enough platforms to treat as a standing warning rather than a one-off complaint. Users also report occasional gaps where an agent misses information sitting in a heavily loaded Drive space, and a recurring desire for more native integrations beyond the current 61 — real limitations for a product whose entire value proposition depends on comprehensive knowledge coverage.
Pricing & Plans
Dust's pricing has genuinely diverged across sources tracking it through 2026 — some show a permanent free tier with a small lifetime credit allocation, others show only a time-limited trial, and tier names (Business vs. Max) differ between trackers — consistent with a platform that has restructured its plans during the year; verify the current structure directly before budgeting.
| Plan | Cena | Co zawiera |
|---|---|---|
| Free / Trial | $0 | Either a one-time 500-credit lifetime allocation or a 14–15 day full-featured trial, depending on which snapshot of the pricing page you're reading — evaluate the actual current terms directly before assuming either |
| Pro | ~$29–30/seat/mo (~$24 annual) | Full platform access: agent builder, 61+ data source connections, multi-model support, roughly 8,000 monthly AI credits per seat — the self-serve entry point for small teams and startups |
| Business / Max | ~$99–150/seat/mo (~$120 annual, per some trackers) | Higher credit allowance (~40,000/seat/month per some sources), designed for teams with heavier agent usage — naming and exact allowance vary by source, verify directly |
| Enterprise | Custom quote (typically 100+ users) | SSO, SCIM, multi-workspace support, custom DPAs, dedicated success management, US/EU hosting choice, and volume-negotiated per-seat rates |
Two buying notes flagged consistently across independent reviews: the Pro tier has a documented per-user data-source storage limit (around 1 GB) that pushes data-heavy teams toward Enterprise faster than the headline price suggests, and per-seat costs at any tier scale linearly with headcount in a way that rewards deliberately limiting seats to actual agent-builders and heavy users rather than licensing an entire department by default.
Key Features & Capabilities
- No-code agent builder — Build custom AI agents grounded in specific company data sources without writing code — the feature that opens agent-building to non-developers across HR, sales, support, and operations, not just engineering teams.
- 61+ native integrations — Direct connections to Slack, Google Drive, Notion, GitHub, and dozens more — the comprehensiveness that determines whether an agent actually knows what your company knows, and the platform's central value proposition.
- Multi-model support — Agents run on GPT, Claude, or Gemini models per task rather than locking into one provider, hedging against any single lab's pricing changes or capability shifts — the same pragmatic flexibility this catalog has praised in Notion's AI architecture.
- Semantic synthesis, not just retrieval — Dust's positioning specifically emphasizes agents that synthesize context across sources rather than merely retrieving and quoting documents — the difference between an assistant that answers a question and one that just finds a paragraph that might contain the answer.
- Enterprise governance stack — SSO, granular per-source permissions, full audit logging, configurable data retention, EU data residency, ISO 27001 and SOC 2 certification, and a no-training agreement with model providers — a genuinely complete compliance package rather than a partial one.
- European sovereignty positioning — Paris headquarters and EU-first data handling make Dust one of the clearer non-American, non-Microsoft options for European enterprises specifically avoiding US jurisdiction — the same structural advantage this catalog credited Mistral for at the model layer, here at the agent-platform layer.
- Collaborative, multiplayer workspace — Agents and their underlying knowledge connections are shared and built collaboratively across a team, building what the platform frames as accumulating institutional memory rather than a personal, single-user tool.
Verdict
Our verdict places Dust in a specific, credible lane: the no-code answer to "we want AI agents that actually know our company," built and priced for teams rather than individuals, with a governance story serious enough to survive procurement.
The clear yes: knowledge-intensive teams — engineering, support, HR, and internal operations — that need agents grounded in real, current company documentation rather than general knowledge, and specifically teams without deep engineering resources to build that grounding themselves. Where Cohere's review in this catalog covered the raw components for teams with developers to assemble them, Dust is the answer for teams that want the assembled result without writing the pipeline. European organizations with genuine data-sovereignty requirements get a real, credible alternative to Microsoft's enterprise Copilot tier — reviewed elsewhere in this catalog — specifically because Dust's EU-first architecture and certification stack were built for exactly that requirement rather than retrofitted onto an American product. And organizations already convinced that scattered internal knowledge is their actual AI bottleneck, rather than raw model capability, are Dust's ideal customer by design.
The honest redirects: small teams or solo users should think hard about whether Dust's per-seat, agent-building complexity solves a problem they actually have — if your real need is a general assistant rather than a knowledge-grounded agent platform, the consumer and prosumer tools this catalog has reviewed elsewhere (Claude, ChatGPT, Gemini) solve that more directly and often more cheaply. Teams evaluating total cost should take the documented 1GB Pro-tier data limit and the credit-scaling-with-seats pattern seriously before committing at a headcount they haven't tested — pilot with a small, deliberately chosen group of actual agent-builders before licensing broadly. Organizations with substantial in-house engineering capacity might find Cohere's raw API components, or building directly on a frontier model's own tooling, offers more control at potentially lower cost than Dust's assembled, per-seat product — the trade is convenience and no-code accessibility against build-it-yourself flexibility. And anyone comparing platforms should budget real time for the documented onboarding friction — multi-day indexing on large document sets and a genuine learning curve for teams new to agent-building are not edge-case complaints but a recurring pattern across independent reviews.
Practical playbook for anyone evaluating a rollout: start the trial or free tier with one specific, well-scoped use case — a single team's documentation, one recurring question pattern — rather than attempting to connect every data source on day one; the multi-day indexing and learning-curve complaints in independent reviews mostly trace back to teams that tried to boil the ocean immediately. Assign a small number of actual agent-builders rather than licensing broadly by default, and let usage patterns justify seat expansion instead of pre-purchasing headcount. Watch the Pro tier's data-source storage limit specifically if your knowledge base is large or growing fast — hitting it mid-year forces an unplanned Enterprise conversation rather than a chosen one. And build the no-training-agreement and data-residency guarantees into your procurement documentation early if compliance sign-off is part of your adoption path, since those are exactly the terms a security review will ask about first.
The competitive map worth drawing explicitly: against Microsoft Copilot's enterprise tier (reviewed elsewhere in this catalog), Dust wins on model flexibility, EU sovereignty, and dedicated agent-building depth, while Copilot wins on default integration for organizations already fully inside Microsoft 365 — the same "already inside the ecosystem you have" logic this catalog has credited Microsoft with elsewhere applies in reverse against Dust for non-Microsoft shops. Against ChatGPT Team or Claude for Work, Dust trades broad general-purpose chat for deep, governed, knowledge-specific agent construction — different products solving adjacent but distinct problems, and the right choice depends entirely on whether your team's actual bottleneck is "we need AI to know our stuff" or "we need AI to be smart in general." Against Cohere specifically, the split is build-versus-buy: Cohere for engineering teams wanting raw, cheap components; Dust for teams wanting the finished product without writing the integration layer themselves.
Weighing it: a genuinely comprehensive integration ecosystem, real no-code accessibility for non-developers, a governance and compliance stack serious enough for regulated industries, credible European sovereignty positioning, and a retention record (zero churn through 2025) that speaks to real, sustained value once teams adopt it — against pricing structures that have shifted enough during 2026 to confuse even dedicated trackers, real onboarding friction on large document sets, per-seat costs that scale meaningfully with headcount, and a value proposition that depends entirely on the quality and coverage of the internal data you actually connect. That lands Dust as a strong, specific recommendation: not a general assistant most individual readers of this catalog need, but a genuinely credible platform for the knowledge-intensive team, especially the European one, whose real AI problem is fragmented institutional knowledge rather than raw model horsepower.
Similar Tools
4.2/5
An enterprise AI platform providing language models for business applications.
Trending
4.2/5
Open-source AI agent framework designed for running and orchestrating models locally.
People Also Ask
What is Dust?
Dust is a no-code AI agent platform built for companies whose real knowledge is buried across tools like Slack, Notion, Google Drive, and GitHub. Rather than functioning as a general chatbot, Dust connects to 61+ of those data sources and lets non-engineers build custom agents grounded in a company's actual internal documentation. Common deployments include onboarding assistants that answer new-hire policy questions, engineering-documentation search bots, and support agents that reference real product docs instead of a generic script. It's model-agnostic, meaning agents can run on GPT, Claude, or Gemini depending on the task, and it's built with an enterprise governance layer (SSO, audit logs, EU data residency) from the ground up. Founded and headquartered in Paris, Dust has positioned itself as much around European data sovereignty as around agent-building itself, which shapes who it's ultimately built for: knowledge-heavy teams, not individual users looking for a general-purpose assistant.
Is Dust worth the cost for a small team?
It depends heavily on team size and how disciplined you are about licensing. Dust's per-seat pricing scales linearly with headcount, and the entry-level Pro tier carries a data-storage cap (around 1GB per user) that data-heavy teams can hit faster than expected, pushing them toward a costlier Enterprise conversation sooner than the sticker price suggests. For a small team, the real question isn't budget alone — it's whether you actually need agent-building infrastructure versus a general assistant. A five-person team with a well-scoped, recurring documentation problem (like repeatedly answering the same onboarding or engineering questions) can get real value from a lean Pro deployment. But a small team without a specific knowledge-grounding problem to solve is often paying for complexity — including a genuine onboarding learning curve — it doesn't need. The practical fix is licensing only actual agent-builders, not a whole department, and validating value on one narrow use case before expanding seats.
Does Dust work well if a company's data is messy or poorly organized?
Reviewers note this is one of Dust's real-world weak points. Because the entire value proposition depends on comprehensive, accurate knowledge coverage, a heavily loaded or poorly structured Drive space, an outdated wiki, or scattered undocumented Slack threads can cause agents to miss information that technically exists in a connected source. Initial indexing of large or messy document sets can also take days rather than minutes, which compounds the problem if the underlying data itself needs cleanup first. This doesn't mean Dust fails with imperfect data — plenty of real companies have messy knowledge bases — but it does mean results are only as good as what you connect. Teams getting the most value tend to start with one well-organized, high-traffic data source (like a maintained handbook or active engineering wiki) rather than connecting everything at once, since a scoped rollout surfaces indexing or coverage gaps early instead of after a full-company launch.
Is Dust better than building agents directly with an AI provider's own tools?
It depends on whether your organization has engineering resources to spare. Dust's core trade-off is convenience versus control: it packages agent-building, integration, and governance into a no-code product that business teams can use without writing a pipeline themselves. Building directly on a frontier model's own tooling, or assembling something with raw components like Cohere's retrieval APIs, can offer more customization and potentially lower cost — but only if you have developers willing to build and maintain the integration and compliance layer yourselves. For teams without that capacity, Dust's assembled product, complete with a no-training agreement, SOC 2 and ISO 27001 certification, and EU data residency already built in, saves meaningful engineering time. The right answer really comes down to whether your bottleneck is a lack of a finished tool or a lack of flexibility — organizations with strong in-house engineering may prefer the build-it-yourself route; those without it generally get to value faster with Dust.
Is Dust free to use?
Not in any lasting sense for real work. The trackers watching Dust's pricing page through 2026 don't even agree on what the free tier includes — some show a one-time 500-credit lifetime allocation, others describe a 14–15 day trial that eventually locks you out entirely. Either way, this is enough to build and test one or two agents against a small data source, not enough to run a team's actual documentation search or onboarding assistant on an ongoing basis. Once you're grounding agents in real Slack history, Notion spaces, or GitHub repos at any meaningful volume, you'll burn through the free allocation quickly and land in Pro territory at roughly $29-30 per seat monthly. Treat the free tier as a proof-of-concept sandbox for deciding whether Dust's agent-building workflow fits your team, not as a viable long-term plan — and confirm the exact current terms directly on Dust's site before assuming either version applies, since the company has visibly restructured its offering more than once during the year.
Does Dust support building agents for customer support use cases?
Yes — it's one of the three concrete use cases reviewers cite most often. A customer-support agent built in Dust can be grounded directly in a company's actual product documentation, help-center articles, and internal wikis rather than running off a generic, hand-written script, which means answers reflect what the product currently does instead of what a support macro said six months ago. Because Dust connects to the real systems where that documentation already lives — Notion, Drive, GitHub, and dozens of other integrations — a support team doesn't need to duplicate or re-upload content into a separate knowledge base just to power the bot. The tradeoff worth planning for: agents are only as good as the underlying data connection, and reviewers note occasional gaps where an agent misses information buried in a heavily loaded Drive space, so a support deployment still needs someone periodically checking that the connected sources are current and complete.
What are the closest alternatives to Dust for enterprise AI agents?
The most relevant alternatives split along two lines. For teams wanting EU-based, non-Microsoft AI with strong sovereignty guarantees, Mistral covers the same territory at the model layer while Dust covers it at the agent-platform layer. For organizations fully committed to Microsoft 365, Microsoft Copilot's enterprise tier is the natural default competitor, trading Dust's model flexibility and dedicated agent-building depth for out-of-the-box integration with tools a company already uses daily. For teams with in-house engineering capacity, Cohere offers the raw retrieval and API components to assemble something similar rather than buying an already-built product. And for organizations whose actual need is a smart general-purpose assistant rather than a knowledge-grounded agent platform, Claude, ChatGPT, and Gemini's team plans solve that more directly and typically more cheaply than Dust's per-seat, agent-building model, which is really built for a narrower, more specific job.
Does Dust keep company data private from AI model providers?
This is one of Dust's stronger, more specific claims — the platform maintains a no-training agreement with the underlying model providers whose GPT, Claude, and Gemini models power its agents, meaning the internal documents, Slack messages, and code an agent reads for context don't get absorbed into anyone's training pipeline. That's paired with EU data residency options, ISO 27001 and SOC 2 certification, granular per-data-source permissions, and full audit logging, which together form a genuinely complete compliance package rather than a marketing checkbox. For regulated industries or IP-sensitive companies, this combination is often the deciding factor over cheaper, less-governed alternatives. It's worth noting these protections apply to how Dust and its model partners handle data in transit and processing — organizations with especially strict requirements should still confirm specific contractual language during procurement, since compliance needs vary meaningfully by industry and jurisdiction.