Kylon vs. Dust: Knowledge layer vs. execution layer

Dust has built one of the more thoughtful enterprise AI platforms. The core idea is right: connect AI to your company's knowledge — Slack, Notion, Google Drive, GitHub, Salesforce — and let agents answer questions, draft responses, and surface insights with real context. SOC 2 Type II certified, GDPR compliant, model-agnostic, and designed for teams, not just individuals.

The no-code agent builder lets domain experts create specialized agents without engineering support. The Spaces system provides granular access control. The RAG pipeline is genuinely good — multiple reviews note that Dust outperforms Microsoft Copilot and ChatGPT at retrieving structured information from internal documents.

We think Dust gets the knowledge layer right. Where we differ is in what comes next.

What Dust does well

Enterprise knowledge integration. Dust connects to 50+ tools and turns scattered company data into organized, searchable context. Agents can access Slack conversations, Notion docs, Google Drive files, GitHub repos, and CRM records — synthesizing information across sources. This is the hardest part of enterprise AI, and Dust does it well.

No-code agent builder. Both technical and non-technical users can create agents through a visual interface. Define instructions, connect knowledge sources, and deploy — without writing code. For enterprises where the people who understand the domain aren't necessarily the people who write software, this accessibility matters.

Granular permissions with Spaces. Public spaces make information company-wide. Private spaces restrict access to specific teams. Administrators control which data sources each agent can see. This is the right way to handle enterprise data governance.

Model flexibility. OpenAI, Anthropic, Google, Mistral, DeepSeek — choose the right model for each task. No vendor lock-in. This pragmatic approach means you're always using the best available model.

Security posture. SOC 2 Type II, GDPR, HIPAA-capable, SSO/SCIM, 365-day audit log retention, zero model training on customer data. For regulated industries, this isn't optional — and Dust takes it seriously.

Table Query. Combining semantic search with SQL for quantitative analysis is a smart capability. Agents that can both reason about text and query structured data cover more ground.

Where the architecture diverges

Dust and Kylon both believe AI should understand your business context. The difference is what the AI does with that understanding.

Knowledge vs. execution

Dust's agents are excellent at knowing things. They search your documents, synthesize information, draft responses, and answer questions. The output is primarily text — answers, summaries, drafts, recommendations.

Kylon's agents know things and do things. They query databases, build dashboards, deploy websites, send emails, run multi-step workflows, monitor KPIs, and coordinate with other agents. The output isn't just text — it's structured data, deployed apps, executed workflows, and real business actions.

The distinction matters in practice. When a sales manager asks "what's our pipeline by region?", a Dust agent can search Salesforce data and write a summary. A Kylon agent can build a live dashboard that updates automatically, alert the team when a deal stalls, and draft the follow-up email — in the same workspace where the team discusses the deal.

Agent as assistant vs. agent as teammate

Dust's agents respond when asked. They're knowledgeable assistants that wait for questions and produce answers. This is valuable — but the interaction model is fundamentally query-response.

Kylon agents are proactive teammates. They monitor channels, surface insights without being asked, run scheduled workflows, and coordinate with other agents on multi-step projects. A marketing agent notices campaign CPA trending up and surfaces it in the channel before anyone checks the dashboard. An ops agent completes a workflow and hands off the next step to the appropriate team member.

The difference is between an AI you query and an AI that participates.

Single agents vs. multi-agent coordination

Dust supports creating multiple agents, but each operates independently. You build a sales agent, a support agent, a data agent — each with its own knowledge sources and instructions. They don't naturally coordinate with each other.

Kylon agents work as a team. Multiple specialized agents operate in the same channels alongside humans. They share context naturally, hand off tasks, review each other's output, and coordinate on multi-step projects. When the marketing agent needs data analysis, it works with the data agent. When the ops agent completes a workflow, it notifies the relevant humans. The coordination is native, not simulated.

Connected data vs. native data

Dust connects to external data sources — it reads from your tools but doesn't own a data layer. Agent-generated insights are returned as text in conversations. There's no persistent, structured storage for the knowledge agents create.

Kylon has a native data layer. Database apps live inside the workspace. Agents read, write, and query structured data directly. When an agent enriches a lead, the data goes into a shared database that other agents and humans can filter, sort, and build on. Knowledge compounds inside the workspace — it doesn't evaporate after the conversation ends.

Platform access

Dust is browser-based. Kylon runs on web, macOS, iOS, and Android. When an agent surfaces an insight or completes a workflow, it reaches you wherever you are — on your phone, on your laptop, in a meeting.

Pricing model

Dust charges per user — €29/user/month for Pro. For a 50-person team, that's ~€1,450/month before you've done anything. This makes sense for knowledge-layer tools where every employee needs access.

Kylon's pricing is designed for teams that work with agents, not just query them. The value scales with what agents produce, not with headcount.

When each makes sense

Choose Dust when your primary need is a knowledge layer — connecting AI to company documents, making information searchable and synthesizable, and deploying no-code agents that answer domain-specific questions. If your team's main pain point is "we have information everywhere but can't find or use it efficiently," Dust's RAG pipeline and enterprise security are genuinely best-in-class.

Choose Kylon when you need agents that go beyond answering questions to executing work. If you want agents that build databases, run workflows, deploy apps, monitor metrics, coordinate across teams, and work proactively alongside humans on web, Mac, iPhone, and Android — Kylon is the execution layer on top of the knowledge layer.

Knowledge is the foundation. Execution is the product.

Both Dust and Kylon start from the same insight: AI without company context is generic and limited. The knowledge layer is foundational — and Dust has built a strong one.

But for most teams, knowing isn't the bottleneck. Doing is. The gap between "we know our pipeline is weak in APAC" and "we've built a tracker, scheduled alerts, and drafted outreach" — that's where work actually happens. Kylon is built for that gap.