# Dust — Platform for building and running AI agents on connected company data, with 20+ models and MCP tools.

> Source: The Agents Index — https://theagentsindex.com/dust (structured, researched, re-verified)
> Facts last verified: 2026-09-09

Dust is a platform for creating AI agents that read a company's own knowledge across tools like Slack, Notion, GitHub and Salesforce. Teams and agents work side by side in shared conversations and workspaces called Pods.

| Fact | Value |
| --- | --- |
| Website | https://dust.tt |
| Price | from €30/month billed monthly, €24/month billed annually |
| Free tier | Yes |

Free option on the Business plan: no credit card, up to 5 users, 3 connectors, 5 Spaces, and 500 credits per free seat as a one-time lifetime allocation that never resets and cannot draw on the workspace pool.

## What it does

**Writes code** · Partly — Agents author React components (Frames) and run code in a sandbox to produce files, but Dust documents no repository-level code authoring or commits.

> A Frame is an interactive file (a React component generated by an agent) that can be stored in the Pod's Files tab.
> — https://docs.dust.tt/docs/user-documentation/pods/frames · Frames page, opening definition

**Debugs and fixes** · Partly — Marketed for debugging by surfacing code context, docs and past issues to the engineer; nothing documents the agent landing the fix itself.

> Streamline debugging within your IDE by surfacing relevant code context, documentation, and recent issues.
> — https://dust.tt/home/solutions/engineering · Your use cases, your way section, Code debugging card

**Reviews code** · Yes — GitHub tools let an agent review pull requests against your internal rules; an admin must first run the OAuth flow and pick the repositories.

> GitHub tools let your agents work with GitHub issues and pull requests. Use them to build agents that file issues when needed, or that review PRs based on internal coding rules.
> — https://docs.dust.tt/docs/user-documentation/agents/tools/github · Overview section, first paragraph

**Whole-codebase work** · Partly — It can ingest a repository's default-branch code for search, refreshed every eight hours, but no documented ability to carry a change across that repository.

> Code Synchronization: Repository code from the default branch
> — https://docs.dust.tt/docs/user-documentation/admins/connections-management/github · Data Synchronization list, Code Synchronization bullet

**Done-for-you service** · No — You buy software and configure the agents yourself; Dust's own people come in as onboarding and customer success on Enterprise, not as a build team.

> Choose self-serve plan for your team, or talk to us about enterprise-ready deployment, governance, and support
> — https://dust.tt/home/pricing · Pricing page header, sub-heading under 'Pricing that scales'

_Not established: Writes tests, Ships it_

## How much it does unattended

**Runs autonomously** · Yes — Agents run unattended on schedules and event triggers and can schedule their own future wake-ups, though high-stake actions still stop to ask the user.

> Multi-agent orchestration & triggers (scheduled + event-driven)
> — https://dust.tt/home/pricing · Compare plans feature by feature table, Features section, Multi-agent orchestration row (Business column ticked)

**Multi-agent** · Yes — One agent can call another as a tool, in the background or as a full handoff, with sub-agent chains capped at a recursion depth of four.

> The run_agent tool enables your agents to run other agent as a tool.
> — https://docs.dust.tt/docs/user-documentation/agents/tools/run-agent · Run agent page, first line

**Agent permissions** · Partly — Tools carry stake levels: high-stake ones such as scheduling a wake-up require user confirmation, and the code sandbox only reaches domains an admin allowlisted.

> Wake-ups scheduling is high-stake and require user confirmation to prevent uncontrolled blow-out usage.
> — https://docs.dust.tt/docs/user-documentation/agents/tools/wake-ups · Restrictions section, final bullet

## Which models it runs on

**Claude** · Yes — Claude is one of the named providers on every plan; the admin tier tables list Claude Haiku, Sonnet and Opus at several reasoning efforts.

> 20+ frontier models — GPT, Claude, Gemini, Mistral, DeepSeek
> — https://dust.tt/home/pricing · Business plan card, feature list, first bullet

**GPT** · Yes — OpenAI GPT models are available on all plans and appear across Dust's Basic, Standard and Premium model access tiers alongside the auto-model lists.

> 20+ frontier models — GPT, Claude, Gemini, Mistral, DeepSeek
> — https://dust.tt/home/pricing · Business plan card, feature list, first bullet

**Open models** · Yes — Open-weight families such as Mistral, DeepSeek, Kimi and GLM are offered, but Dust serves them itself through providers like Fireworks rather than your own weights.

> Build custom agents with 20+ frontier and open-source models from OpenAI, Anthropic, Google, Mistral, and more.
> — https://dust.tt/home/product · Model flexibility without lock-in section

**Model choice** · Yes — You pick the model per agent or per message, or delegate to a curated auto model; nothing is reserved for a higher plan, only priced in credits.

> You choose the model per agent. No model is locked behind a higher plan, though higher-capability models may consume more credits per message.
> — https://dust.tt/home/pricing · Pricing FAQ, 'Which AI models are included?'

## Where you use it

**In your editor** · Partly — No native editor plugin ships; an IDE reaches Dust as an MCP client, and the CLI can install a skill so a local coding agent calls Dust agents.

> This lets tools such as IDEs and AI assistants call Dust to search workspace knowledge, work with conversations, and access files or Pods through the Model Context Protocol.
> — https://docs.dust.tt/docs/user-documentation/agents/integrations/dust-mcp-server · Dust MCP Server page, introduction

**On the command line** · Yes — A published npm CLI opens an interactive terminal chat with filesystem access, and takes single non-interactive messages that return JSON on stdout.

> When no command is provided, the chat command runs by default, opening an interactive chat with @Dust with access to your local filesystem.
> — https://docs.dust.tt/docs/developer-platform/dust-cli/dust-cli · Usage section

**In your pipeline** · Partly — The CLI authenticates headlessly for pipelines and agents can review PRs, but Dust publishes no first-party CI action; you wire it in yourself.

> For automated workflows, CI/CD pipelines, or environments without interactive terminals, the CLI supports headless authentication.
> — https://docs.dust.tt/docs/developer-platform/dust-cli/dust-cli · Headless authentication section, opening line

**In a browser** · Yes — The primary surface is a hosted web app of conversations and Pods, complemented by a browser extension, Slack, Teams and Zendesk entry points.

> Dust acts as a replacement for ChatGPT, Gemini and other similar tools. Simply call one of the generic agents (ex: @gpt) to use it.
> — https://docs.dust.tt/docs/user-documentation/getting-started/intro-to-dust · Ask your first question, 'Talk to the best LLMs'

## Whose machine it runs on

**Self-hosted** · Partly — The isolation option Dust sells is a dedicated single-tenant deployment it runs on Enterprise; the MIT-licensed source exists but no self-hosting guide is published.

> Single-tenant deployment
> — https://dust.tt/home/pricing · Compare plans table, Security & admin section, Single-tenant deployment row (Enterprise only; Business shows a dash)

**Open source** · Yes — The platform's code sits in a public repository the vendor links from its own footer, under a permissive MIT licence covering modification and redistribution.

> Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction
> — https://github.com/dust-tt/dust/blob/main/LICENSE · LICENSE file, second paragraph

## What it costs to run

**How it meters** · Yes — Billing is per seat per month, and each seat carries a credit allowance consumed by token use plus a fixed action charge of 0, 1 or 3 credits per tool call.

> Credits are the unit of usage on Dust. Every AI interaction (sending a message, running an agent, using a tool) consumes credits.
> — https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/credits · Credits page, opening paragraph

**Free tier** · Yes — A standing free option needs no card, but caps you at five users, three connectors and a one-time 500-credit allowance that never renews.

> Business: available with a free option (no credit card required, up to 5 users, 3 connector, 5 Spaces) or with paid Pro and Max seats for more capacity
> — https://docs.dust.tt/docs/user-documentation/admins/billing/subscriptions-and-payments · Plans section, Business bullet

**API access** · Yes — A documented REST API, JavaScript SDK and CLI drive agents; Business gets the Conversation API only, the Data Source API is added on Enterprise.

> The developer platform lets you build on Dust programmatically through the Dust API, the JavaScript SDK, and the Dust CLI.
> — https://docs.dust.tt/docs/developer-platform/overview/developer-platform · Developer platform page, third paragraph

**MCP server** · Yes — Dust runs a remote MCP server with OAuth so another agent can list agents, search Spaces and post conversations as the authenticated user.

> Dust can expose your workspace as a remote MCP Server, so compatible MCP clients can use Dust where you already work.
> — https://docs.dust.tt/docs/user-documentation/agents/integrations/dust-mcp-server · Dust MCP Server page, first line

_Not established: Bring your own key_

## Buying it for a team

**A company can buy it** · Yes — Sold to companies on both plans: Business self-serve with VAT-registered business checkout up to 100 seats, Enterprise via sales with custom terms.

> Enterprise: unlimited users and connectors, unlimited Spaces, SSO, SCIM, audit logs, and advanced credit controls. Contact sales to get started.
> — https://docs.dust.tt/docs/user-documentation/admins/billing/subscriptions-and-payments · Plans section, Enterprise bullet

**Seat model** · Yes — Per seat with three types (Free, Pro, Max) mixable in one workspace; the free option stops at five users and Business at 100, Enterprise is annual only.

> On the Business plan, seats can be billed monthly or annually, and both billing cycles can coexist within the same workspace. Annual seats are billed at a lower monthly rate.
> — https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/seat-management · Monthly vs. annual seats section

**Pooled budget** · Yes — Each paid seat has its own credits and falls back to a shared workspace pool; Enterprise Pooled drops personal allowances entirely for one balance.

> The workspace credit pool is a shared credit reserve that Pro and Max seat members can draw from once their individual credits are exhausted.
> — https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/credit-management · Workspace credit pool section

**Admin controls** · Yes — Admins cap which models and reasoning efforts each member may pick, set per-member spend limits and programmatic caps, and allowlist sandbox domains.

> As an administrator, you can cap how far up the model list each member can go.
> — https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/model-access-tiers · Model access tiers, introduction

**Audit log** · Partly — A searchable, exportable log distinguishing human from agent actions exists, but only on Enterprise; the Business column of the pricing table shows a dash.

> Enterprise feature. Audit Logs are only available to workspaces on the Enterprise plan.
> — https://docs.dust.tt/docs/user-documentation/admins/audit-logs/audit-logs · Audit Logs page, callout above Accessing Audit Logs

**Single sign-on** · Yes — SAML SSO with Okta, Entra ID and JumpCloud, enforceable workspace-wide; on Business it is on demand from five seats, and SCIM stays Enterprise-only.

> Dust supports Single Sign-On (SSO) to manage your team's access to Dust through your existing Identity Provider (IdP), so authentication is managed in one place.
> — https://docs.dust.tt/docs/user-documentation/admins/admin-governance/single-sign-on-sso/single-sign-on-sso · Overview section

## What happens to your code

**Opt out of training** · Yes — No toggle to find: the vendor contractually bars its model providers from training on customer data and states zero retention on their side.

> Foundational Model Providers (OpenAI, Anthropic, Mistral, Google, Fireworks) are prohibited from using any customer and personal data for model training.
> — https://dust.tt/home/platform-privacy · Section 6, 'Our service providers'

**Data residency** · Yes — You choose US or EU hosting on both plans, with a separate EU endpoint; the company itself is Paris-based and data may still be processed in the US.

> Host in the EU or US to meet your regulatory needs.
> — https://dust.tt/home/security · Ingest data on your terms section, Regional Hosting item, body line

**Getting out** · Partly — Analytics, audit logs and agent configs export as CSV, JSON or YAML, but credits never survive the period and cancelling deletes every connection.

> Credits are not transferable between users, do not roll over to the next billing period, and are not refundable.
> — https://docs.dust.tt/docs/user-documentation/admins/usage-seats-and-credits/credits · What is a credit? section, second paragraph

**Certifications** · Yes — SOC 2 Type II and GDPR are claimed on both plans with a public trust centre; HIPAA is framed as enabled rather than certified, and no ISO 27001 claim appears.

> GDPR Compliant & SOC2 Type II Certified. Enables HIPAA compliance.
> — https://dust.tt/home/security · Designed for enterprises section, badge line

## Provenance

Every fact above comes from the vendor. Nothing here is independently corroborated yet.

## Coverage elsewhere

None recorded. Every fact above comes from the vendor and is not independently corroborated yet.

## What changed

- 2026-09-09 — Capabilities · Commercial terms · Fit check passed · Free tier · Price

## Correcting this record

Anything marked "Not established" is a question we have not answered — it is not a No. If you represent Dust, or anything here is wrong or out of date, you can supply it with a source: the operations this site accepts from an agent are listed at https://theagentsindex.com/.well-known/agents.json — that catalogue needs no key to read, and each entry states its own auth.

Supplying evidence is free. Nothing that can be bought here changes what a field says, whether a gap is closed, or how this record is scored.