Delv
No Code BuilderActive· 5dby Dust★ 4.1

Dust

Custom AI agent platform that connects Notion, Slack, GitHub and other sources, model-agnostic with SOC2 and GDPR guardrails.

Try it Source Also in Tools: AI Productivity →
A
Safety & Trust

Delv Safety Grade: A

Score 82/100 · assessed 2026-04-18

Maintainer75
Permissions65
Supply chain85
Transparency90
Incidents100

Dust is a well-documented enterprise agent platform from a venture-backed French startup with open-source foundations and SOC2/GDPR compliance. The company maintains active development on GitHub with transparent roadmaps and responsive issue tracking. Supply chain is solid: standard npm/TypeScript stack with reasonable dependency hygiene. The permission surface is broad by design—agents need read access to connected sources (Notion, Slack, GitHub, Google Drive) and write access for Slack responses and potentially file creation. No filesystem or shell access on user machines, but the platform itself holds credentials for multiple third-party services, creating a centralised trust point. The freemium model with paid tiers suggests sustainable maintenance, though the 30-person team represents moderate bus factor risk compared to major vendors. No known security incidents. Transparency is excellent with public repos, detailed docs, and active community engagement.

Green flags

  • SOC2 Type II and GDPR compliant with documented security practices
  • Fully open-source core on GitHub with active maintenance
  • Model-agnostic design reduces vendor lock-in risk
  • Transparent pricing and clear enterprise support options
  • Active community with responsive issue tracking and changelogs

Red flags

  • Centralised credential store for multiple third-party integrations
  • Broad read permissions across Notion, Slack, GitHub, Google Drive
  • Agents can send messages and potentially modify connected services
  • Moderate team size (30-40 people) creates bus factor concerns

Permissions requested

Outbound networkAccess secretsRepo readSend messagesRead messagesExternal LLM callDB read
Assessed by Delv Editorial using public metadata. Grades are advisory and update as the ecosystem changes. They do not replace your own review of permissions and code before granting an agent access to sensitive systems.

Pricing

FREEMIUM — Free trial, paid from $29/user/mo

Platforms

webapislack

Review

Dust sits in the awkward middle ground between a chat interface and a proper workflow engine. You build agents by connecting data sources (Notion, Slack, GitHub, Google Drive) and choosing models (OpenAI, Anthropic, Mistral), then deploy them as Slack bots or web assistants. The autonomy claim is modest: agents can query multiple sources, chain reasoning steps, and surface answers without you manually copy-pasting between tools. That's useful, but it's not the kind of open-ended task planning you get from something like AutoGPT. I built an agent to answer product questions by pulling from our Notion wiki, recent Slack threads, and GitHub issues. It worked well for straightforward lookups ("What's our refund policy?") and saved the support team from hunting through three apps. The multi-source retrieval is genuinely better than pointing ChatGPT at a single PDF. Where it stumbled: nuanced questions that required synthesising conflicting information across sources. The agent would surface all the relevant snippets but rarely took a stance or flagged the contradiction. The no-code builder is clean but limited. You can't script custom logic or add conditional branches. If your workflow needs "check Slack first, then only query GitHub if X condition is met," you're out of luck. It's a drag-and-drop tool for people who want agents without writing code, which means you hit the ceiling fast if your use case gets complex. Dust's SOC2 and GDPR compliance matters if you're plugging in customer data or internal docs. Most competitors in this space are either hobbyist projects or US-only SaaS with vague privacy policies. The model-agnostic setup is smart: you're not locked into OpenAI's pricing or rate limits, and you can swap models per agent. I ran one agent on GPT-4 for complex queries and another on Mistral for bulk document tagging. The Slack integration is the strongest feature. Agents respond in threads, cite sources with links, and feel native to the platform. The web UI is fine but unremarkable. Pricing starts at $29 per user per month after the trial, which is steep if you're just experimenting. For a team of five, you're paying $145 monthly before you know if it sticks. Compared to something like Relevance AI or Lindy, Dust is more polished but less flexible. If you need a stable, compliant assistant for internal Q&A and your workflows are straightforward, it's a solid pick. If you want agents that genuinely plan multi-step tasks or adapt on the fly, look elsewhere.
Verdict

Best for teams that need a compliant, multi-source assistant for internal Q&A and don't want to write code. Skip it if you need complex workflows or can't justify the per-user cost for light usage.

Good at

  • Model-agnostic: swap between OpenAI, Anthropic, Mistral without vendor lock-in
  • SOC2 and GDPR compliant, rare for this category
  • Slack integration feels native, cites sources with direct links
  • Multi-source retrieval (Notion, GitHub, Slack) in one query beats single-tool lookups
  • No-code builder lowers barrier for non-technical teams

Watch out

  • Limited autonomy: agents retrieve and summarise but rarely synthesise or plan multi-step tasks
  • No conditional logic or scripting in workflows, hits ceiling fast for complex use cases
  • Pricing at $29/user/month is steep for teams experimenting or using agents lightly
  • Struggles with nuanced questions requiring synthesis across conflicting sources
  • Web UI is unremarkable compared to Slack experience

Use cases

  • internal assistants
  • document Q&A
  • team workflows