Comparison
Syniaps or Dust?
A 163-person Paris scale-up, $60m raised, an agent workshop sold per seat and plugged into Slack and Notion
What Dust sells: "Multiplayer AI for human-agent collaboration": a shared space where humans and agents work on the same context, with 20+ models, 70+ connectors and enterprise governance (SSO, SCIM, audit).
Syniaps
4.0 / 5
Dust
3.5 / 5
Surveyed on 15 September 2026 from the publisher's public pages, cited at the foot of this page. Offers change: check with the publisher before you decide.
Dust at a glance
- Who
- A Paris scale-up, Sequoia, $60m
- Dust (Permutation Labs SAS, Paris) was founded in February 2023 by Stanislas Polu (ex-Stripe, ex-OpenAI) and Gabriel Hubert (ex-Stripe); €16m in 2024 then a $40m Series B in May 2026 (Sequoia, Abstract, Snowflake, Datadog), 163 employees, ARR up from $1m to $20m in two years, not yet profitable. SOC 2 Type II, GDPR, HIPAA available, 3,000 customer organisations (Doctolib, Qonto, Alan, Pennylane, Datadog, Vanta, Cursor), 300,000 agents deployed, and an MIT-licensed open source core (1.5k stars).
- What
- An agent builder wired to your data
- A web workspace (plus Slack, a Chrome extension, an API and a CLI) where you create "@name" agents from instructions, a model, tools and synced data sources (Drive, Notion, Confluence, Slack, GitHub, Microsoft, Salesforce, Zendesk, Snowflake, BigQuery, Gong, Intercom). Building blocks: web search and browsing in read-only mode, "Computer" (code and files in a sandbox with no internet), file and image generation, per-agent memory, reusable Skills, MCP servers, scheduled triggers and webhooks (November 2025).
- How it gets set up
- You write, choose, test and publish
- A self-serve account (Google, GitHub, email, SSO), then an admin connects the sources and selects the folders, channels and pages to sync (a maximum of 3 connectors on Business). Each agent is built by hand: write its instructions (a "Sidekick" helps), pick the model, tools and sources, describe each source, test it in the preview, give it a handle and publish; automations are triggers described in natural language inside the builder. Dust supplies a six-phase "rollout guide" (strategy, a community of builders, admin configuration, use cases, training, measurement): this is a deployment programme, not a standing start.
- Which models
- 20+ models, chosen per agent
- GPT, Claude (Opus, Sonnet, Haiku), Gemini, Mistral and DeepSeek: the builder picks the model for each agent, with no automatic escalation mode. Inference goes to OpenAI, Anthropic, Google, Mistral or Fireworks under Zero Data Retention contracts that forbid training; the platform runs on Google Cloud with a choice of EU or US residency.
- Price
- Free, then €24 to €150 per seat
- Free: 500 lifetime credits; Pro: €24 per seat per month billed annually (€30 monthly) with 8,000 credits per seat per month; Max: €120 annually (€150 monthly) with 40,000 credits; Enterprise by quote (unlimited connectors and MCP, single tenant, a CSM, MSA/DPA). The credit is the real unit: "Credits are charged per message, based on the model used and the actions performed", so AI is included but overage is billed on top ($0.01 per credit for programmatic use).
- Who it is for
- Scale-ups and mid-market companies on Slack
- Companies of 50 to several thousand people who live in Slack, Notion and a CRM, with internal "AI operators" ready to build and maintain agents for support, sales, engineering and marketing. It is genuinely strong at French and American tech scale-ups (Qonto, Alan, Malt, Clay, Vanta) that want to govern AI across a whole organisation.
Point by point
| Criterion | Dust | Syniaps | Advantage |
|---|---|---|---|
| Intelligence | 20+ models (GPT, Claude, Gemini, Mistral, DeepSeek) chosen per agent, with no escalation mode | Claude, ChatGPT or the model of your choice, with Standard / In-depth / Maximum modes | Level |
| "Execute" | Code and files in a sandbox, the web read-only, writes through connectors and MCP; nothing for a site without an API | Files, code, terminal, a browser driven with cookies, long-running work, hosted apps | Syniaps |
| Knowledge | Sources synced from what the admin selected, semantic search; 3 connectors on Business | Synapses built and maintained on their own from documents, email and a cloud mirror | Syniaps |
| Setup | Every agent is written, tested and published; a six-phase rollout | A short wizard, then the agent proposes and the human confirms | Syniaps |
| Learning | Memory per agent and per user, with Skills written by builders | Corrections and preferences retained, and what it learns proposed for approval | Syniaps |
| Sovereignty | SOC 2 Type II, GDPR, EU residency on GCP, single tenant on Enterprise, ZDR at the providers | A dedicated OVH machine in France per customer, a sandbox per user, AI processed in Europe if you choose, but no certification | Level |
| Entry and price | Free and self-serve, Pro at €24 per seat, AI included in the credits, overage billed | A free 7-day trial within minutes, a card required, from €99 excl. tax a month, AI included | The alternative |
| Scope | 3,000 organisations, 70+ connectors, Slack, Chrome, API, MCP, 163 people | Deeper execution per operator, eleven expert teams, a narrower footprint | The alternative |
What each one does better
Where Syniaps wins
It really acts
Terminal, a browser driven on sites without an API, long-running work, hosted apps: Dust stops at its connectors and a sandbox with no internet.
Nothing to build
With Dust every agent is written, tested and published; Syniaps starts from a short wizard and proposes the automations itself.
Knowledge maintained on its own
Synapses build themselves from documents, email and a full cloud mirror, where Dust indexes whatever an admin ticked, with three connectors on Business.
A dedicated machine in France
Each customer has their own OVH machine and a sandbox per user; Dust is multi-tenant on GCP except on Enterprise.
Learning that is proposed
Syniaps retains corrections and preferences and proposes what it has learnt for approval; Dust's memory is private per agent and per user.
Where Dust wins
Immediate, free entry
A self-serve account in two minutes, 500 credits given for life, then €24 per seat with the AI included: a far lower way in than Syniaps's €99 excl. tax a month.
Proven enterprise governance
SOC 2 Type II, HIPAA available, SSO, SCIM, groups, private Spaces and audit: what Syniaps has not yet certified.
Multiplayer by design
Agents shared by @handle, common Skills, Spaces per group and thread-aware Slack: the whole organisation works on the same context.
Breadth of connectors
70+ managed connectors (CRM, support, data warehouses, Gong, Intercom) plus MCP, Zapier, Make, n8n, an API and a CLI.
Model choice and maturity
20+ models including Gemini and DeepSeek, an open source core, 3,000 organisations and 163 people to hold it all up.
Two caveats on its promise
- "Working while you sleep" is still a builder: triggers are personal, only the editor sees the runs, and every agent must first be written, equipped and published by a builder or an admin.
- The per-seat price hides the real unit, the credit: opaque consumption per message and per model, overage billed, and Business capped at 3 connectors, with single tenancy and unlimited use reserved for Enterprise by quote.
The ten criteria, scored and justified
Intelligence
Frontier models (Claude, GPT, Gemini) and ways to step up a gear
Syniaps 5/5
Claude Sonnet 5, Opus 5 and Fable 5, GPT through Codex, or the model of your choice; Standard, In-depth and Maximum modes.
Dust 4/5
20+ frontier models (GPT, Claude Opus/Sonnet/Haiku, Gemini, Mistral, DeepSeek) chosen per agent, but no documented automatic step-up mode.
Execution power
Actually acts: files, code, browser, sites without an API, long-running work
Syniaps 5/5
Files, code, terminal, a browser for sites without an API, long jobs in the background, hosted apps.
Dust 3/5
Computer runs code and transforms files without internet access, the web is read-only, and writes go through connectors and MCP; no driven browser, nothing for a site without an API, and no long-running background work.
Company knowledge
A base built and maintained on its own from documents, email and cloud storage
Syniaps 4/5
Synapses built by the agent, OCR, a full cloud mirror; still young on very large volumes.
Dust 3/5
Connectors sync automatically within minutes, but the admin picks the folders and channels, Business caps at 3 connectors, and no base is built or maintained by the agent.
Zero setup
Nothing to brief, trigger or tick; the tool proposes, the human confirms
Syniaps 4/5
A short wizard, and the agent proposes its automations in conversation; site connections still have to be made once.
Dust 2/5
Every agent is written, equipped, tested and published by hand, and Dust recommends a six-phase rollout with a community of builders.
Continuous learning
Remembers corrections and preferences without being asked to
Syniaps 4/5
Corrections, skills and preferences are retained and proposed for approval; no learning across customers yet.
Dust 3/5
The Agent Memory tool decides on its own what to keep but stays private per user and per agent; Skills are written and improved by humans.
Autonomy
Missions and automations server-side, computer switched off, human approval on sensitive moves
Syniaps 4/5
Missions and scheduled actions run server-side with your computer off, with approval cards; mission ceilings still need tuning.
Dust 3/5
Scheduled triggers and webhooks (GitHub, Jira, Zendesk) run server-side and write into the tools, but they are personal, audited after the fact, with no documented human approval on sensitive moves.
Integrations and connections
Everyday tools, MCP, cloud storage, email, sites without an API
Syniaps 4/5
Email, cloud storage, meetings, WhatsApp, LinkedIn, MCP, sites without an API; fewer native connectors than a Zapier.
Dust 4/5
70+ connectors, MCP servers added by the admin (OAuth, bearer), Slack, a Chrome extension, Zapier/Make/n8n, an API and a CLI; nothing for sites without an API.
Sovereignty and security
EU hosting, per-customer isolation, EU or on-premise inference, certifications
Syniaps 3/5
A dedicated machine in France (OVH), strong per-user isolation and AI processed in Europe if you choose; no on-premise option and no certification.
Dust 4/5
SOC 2 Type II, GDPR, HIPAA available, EU residency on Google Cloud and ZDR at all five model providers, but multi-tenant by default (single tenant on Enterprise) and inference outside the EU.
Teamwork
Shared spaces, roles, delegation, company-wide memory
Syniaps 4/5
Company sources connected once, shared synapses, role profiles, shared skills and conversations, shared mailboxes in missions; no SSO and no fine-grained roles.
Dust 5/5
Spaces, SCIM groups, admin/manager/member roles, additive permissions, shared agents and Skills, Company Data: this is the heart of the "multiplayer" product.
Price and access
Total cost (tool + AI + machine) against what you get, and how easily you start
Syniaps 3/5
A dedicated machine, AI included and a 7-day trial that opens within minutes, from €299 excl. tax a month; a card is required for the trial and there is no free tier.
Dust 4/5
Free for life (500 credits), self-serve, €24 to €150 per seat with AI included, but opaque credits, billed overage and Enterprise by quote for single tenancy.
Verdict
The per-seat agent workshop against the intelligence that acts on a dedicated machine
Dust
Dust equips teams that want to build and govern their own agents inside Slack and Notion, with the AI included in the seat.
Natural fit: scale-ups and mid-market companies of 50 to 5,000 people, a Slack culture, internal builders and an IT department that demands SOC 2 and SSO.
Syniaps
Syniaps delivers an intelligence that already knows the company and runs missions end to end on a dedicated machine, with nothing to build.
Natural fit: companies with no internal builder that want missions done (sites without an API, files, apps) rather than an agent workshop.
Do not fight on Dust's ground (workshop, seats, connectors) but on what it does not do: act without an API, know on its own, ask you to build nothing.
Hold the "nothing to brief, nothing to tick" promise as the structural difference, and document the isolation of a dedicated machine while waiting for a certification. The 7-day trial that opens within minutes and the AI included in every plan have made the way in lighter, but Dust keeps the edge at the door with its free tier and its €24 per seat.
Public sources (25)
- https://dust.tt
- https://dust.tt/home/pricing
- https://dust.tt/home/security
- https://dust.tt/home/platform-privacy
- https://dust.tt/blog/series-b-multiplayer-ai
- https://dust.tt/blog/introducing-triggers-your-agents-working-while-you-sleep
- https://docs.dust.tt/docs/user-documentation/agents/create-your-first-agent.md
- https://docs.dust.tt/docs/user-documentation/agents/tools/index.md
- https://docs.dust.tt/docs/user-documentation/agents/tools/computer.md
- https://docs.dust.tt/docs/user-documentation/agents/tools/web-search-and-browse.md
- https://docs.dust.tt/docs/user-documentation/agents/tools/agent-memory.md
- https://docs.dust.tt/docs/user-documentation/agents/skills/skills-overview.md
- https://docs.dust.tt/docs/user-documentation/agents/triggers/schedules.md
- https://docs.dust.tt/docs/user-documentation/agents/triggers/credits-usage.md
- https://docs.dust.tt/docs/user-documentation/data-sources/connections.md
- https://docs.dust.tt/docs/user-documentation/admins/tools-management/adding-an-mcp-server.md
- https://docs.dust.tt/docs/user-documentation/admins/admin-governance/workspace-governance-roles-groups-and-permissions.md
- https://docs.dust.tt/docs/user-documentation/getting-started/dust-rollout-guide/index.md
- https://github.com/dust-tt/dust
- https://www.maddyness.com/2026/05/18/ia-dust-leve-40-millions-de-dollars-pour-accelerer-sa-croissance-aux-etats-unis/
- https://www.blogdumoderateur.com/histoire-dust-pepite-francaise-ia-racontee-par-fondateur/
- https://automationatlas.io/answers/dust-pricing-explained-2026
- https://www.g2.com/products/dust-dust/reviews
- https://www.capterra.com/p/10019510/Dust/
- https://signin.dust.tt
Dust is a trademark of its publisher, cited here for the purpose of objective comparison. Syniaps is neither affiliated with nor a partner of that publisher. Found an error in this comparison? Write to us at [email protected] and we will correct it.