How Teams Connect Notion, Slack, and Google Drive to Dust to Build Custom Internal AI Assistants

Enterprise teams use Dust to break down information silos across Google Drive, Notion, and Slack, turning fragmented documents and daily conversations into responsive internal AI assistants. Rather than building custom retrieval-augmented generation (RAG) pipelines from scratch, organizations connect their existing software tools to Dust through managed integrations, organize the synced data into secure Spaces, and configure specialized agents equipped with targeted instructions and model capabilities. Employees can then query these assistants directly from Slack or a central web workspace to retrieve accurate, source-grounded answers without compromising confidential organizational data.

Understanding Dust Architecture: Connections, Spaces, and Permissions

To connect enterprise tools securely, administrators must understand how Dust structures data ingestion and access governance. Dust separates the ingestion mechanism from user authorization through three distinct layers: Connections, Spaces, and Assistants.

Connections vs. Spaces

A Connection represents the technical integration between Dust and a third-party software platform (such as Google Workspace, Notion, or Slack). Configured by a workspace administrator, a connection handles API authentication, webhooks, and continuous data synchronization.

A Space is a logical container within Dust that holds synced data and documents. Instead of dumping all company information into a single unstructured repository, administrators create separate Spaces to mirror organizational boundaries. Dust provides two types of Spaces:

• Open Spaces: Accessible to all members of your Dust workspace. These are ideal for company-wide materials like handbooks, product documentation, brand guidelines, and public announcements.
• Restricted Spaces: Accessible only to designated team members or security groups synced from an Identity Provider (IdP). These are reserved for sensitive data, such as executive planning, legal agreements, human resources records, or financial models.

The Dual-Layer Permission Model

Dust uses a dual-layer security model to prevent unauthorized data exposure across automated workflows:

1. Agent-to-Data Access: When a creator builds an assistant, they select which Spaces the assistant can search. The assistant inherits access to all data stored within those assigned Spaces.
2. User-to-Agent Access: By default, a team member can only interact with an assistant if that user has permission to access all Spaces attached to the assistant. If an assistant references both an Open Space and a Restricted HR Space, an employee outside the HR group cannot converse with that assistant, preventing inadvertent data leaks.

Service Account Best Practice

When establishing connections, avoid using personal employee accounts. If an employee leaves the company or their credentials expire, connected data pipelines can break unexpectedly. Instead, provision a dedicated administrative service account—such as dust-admin@yourcompany.com—with appropriate workspace-level viewing rights across Google Workspace, Notion, and Slack before authorizing connections in Dust.

Connecting Core Enterprise Data Sources

Configuring integrations in Dust requires administrative credentials in both Dust and the target application. From the Dust dashboard, navigate to Spaces > Connections to initiate the setup for each platform.

Google Drive

Connecting Google Drive allows assistants to search and extract context from Google Docs, Sheets, Slides, and uploaded PDF files.

1. Authentication: In the Connections menu, select Google Drive and click Connect. Log in using your dedicated admin account and grant Dust read-only access to your Google Workspace drive contents.
2. Scoping Folders and Drives: Dust allows granular folder selection. You can select entire Shared Drives (recommended for collaborative departments) or specific nested folders within My Drive. Avoid syncing personal user drives or unorganized root folders.
3. Indexing and Sync Time: Once authorized, Dust begins indexing file contents and generating vector embeddings. While text-heavy Google Docs process quickly, large drives containing tens of thousands of files or extensive PDFs can require several hours to complete their initial synchronization.
4. Assigning to Spaces: After data sync completes, navigate to your target Space (e.g., "Engineering Knowledge" or "Company General") and add the relevant Google Drive folders to that Space.

Notion

Notion integration enables assistants to query team wikis, project trackers, product requirement documents (PRDs), and structured databases.

1. Authorization: Select Notion from the Connections directory. The Notion authorization screen will prompt you to select the Notion workspace and approve the integration.
2. Page and Database Selection: Notion's permission structure requires explicitly granting the Dust integration access to top-level pages. Select the parent pages or team databases you want to sync. Dust automatically indexes child sub-pages nested under authorized parent pages.
3. Syncing Relational Databases: Dust indexes database properties and page body content. For complex databases with linked relations, ensure all referenced databases are shared with the Dust integration to maintain data integrity in assistant answers.
4. Continuous Updates: Dust listens for page edits and database updates, updating its internal search index incrementally as your team edits documentation in Notion.

Slack

Integrating Slack brings conversational context, troubleshooting threads, and announcement histories into your internal AI ecosystem.

1. Installing the Dust Slack App: Select Slack in the Connections view and authorize the Dust Slack app for your organization's Slack workspace.
2. Scoping Public Channels: During setup, choose which public channels Dust should index. Prioritize channels with persistent knowledge value, such as #announcements, #product-updates, #customer-feedback, or #engineering-help. Exclude high-noise channels like #random or #social.
3. Handling Private Channels: For security reasons, Dust does not automatically index private channels. To include knowledge from a private project channel, an administrator or channel owner must manually invite the bot into the channel by typing /invite @Dust.
4. Message Threading Context: Dust captures conversation threads so assistants understand the full context of questions and answers rather than evaluating isolated messages in isolation.

Step-by-Step Guide: Building Custom Internal Assistants

Once your data sources are connected and organized into Spaces, you can assemble custom AI assistants tailored to specific business roles.

Step 1: Define Identity and Role

Navigate to the Dust home screen and select Create Agent (or Assistant). Establish clear identifying attributes:

• Name: Give the assistant an intuitive name that reflects its specialization (e.g., SalesEnablement, TechDocsBot, PeopleOpsGuide).
• Handle: Set a short handle for mentions (e.g., @sales-help).
• Description: Provide a precise, single-sentence summary of what the assistant does. Dust uses this description to route generic questions to the most qualified agent when users query the default @dust orchestrator.

Step 2: Choose the Underlying Foundation Model

Dust provides model flexibility, allowing creators to select the best foundation model for the assistant's complexity and latency requirements:

Step 3: Author System Instructions and Guardrails

The instructions field governs the assistant's behavior, tone, reasoning methodology, and fallback responses. Effective instructions follow a structured prompt template:

You are the internal People Operations Assistant for Acme Corp.
Your role is to help employees understand company policies, benefits, and workplace guidelines based exclusively on verified internal documents.

Guidelines:
1. Always base your answers on information retrieved from the connected Notion "People Wiki" and Google Drive "HR Policies" folders.
2. If a policy document does not contain the answer, state clearly: "I cannot find documentation on this topic in our verified HR spaces. Please consult the People team directly in #ask-people."
3. Never speculate or fabricate company rules, leave allowances, or compensation details.
4. When explaining multi-step procedures (e.g., submitting expense reports or requesting parental leave), provide clear numbered steps and cite the title of the source document.

Step 4: Attach Knowledge and Configure Retrieval

In the Tools & Knowledge section of the assistant builder:

1. Select Data Sources: Click Add Knowledge and choose the Spaces or specific connected folders relevant to the assistant. For example, a customer support assistant should be linked to the Zendesk knowledge base, Notion troubleshooting pages, and the #support-resolved Slack channel.
2. Add Source Descriptions: Provide a brief note explaining what each attached source contains. This helps the assistant decide which knowledge base to query first when answering multi-part questions.
3. Configure Additional Tools: If desired, enable auxiliary capabilities such as real-time web browsing, table computation, or custom API actions.

Step 5: Test and Iterate

Use the interactive preview pane on the right side of the builder to test the assistant before publishing:

• Test Standard Inquiries: Ask common questions to verify that the assistant retrieves the correct documents and summarizes them accurately.
• Test Edge Cases and Negative Queries: Ask questions about topics outside the assistant's scope to ensure its guardrails prevent hallucination and properly trigger fallback notices.
• Inspect Retrieved Chunks: Check the citation references beneath the preview response to verify that Dust is pulling relevant document chunks from Notion, Drive, or Slack.

Deploying Assistants to End Users Across Workplace Tools

Building assistants in Dust is only half the process; widespread adoption depends on meeting team members in the applications they already use.

Slack Integration

The Dust Slack app brings assistants directly into daily communication channels:

• Direct In-Channel Mentions: In any channel where the Dust app is installed, users can summon a specific assistant by typing @dust +AssistantName followed by their question (e.g., @dust +TechDocsBot How do we configure local staging environments?).
• Threaded Answers: Dust posts its response directly in the message thread, preserving channel cleanliness while keeping context accessible to all participants.
• Default Company Assistant: Querying @dust without specifying an agent invokes the default company assistant, which searches across all open organizational documentation.

Dust Web Workspace and Multi-Agent Collaboration

For in-depth research, employees can use the Dust web interface (dust.tt):

• Multi-Agent Conversations: Users can invoke multiple assistants in a single discussion thread. For instance, a product manager can mention @TechDocsBot to explain an architecture constraint and then call @SalesEnablement in the next turn to translate that technical detail into a customer-facing benefit.
• Source Citations: Every answer in the web workspace includes clickable citations that direct users to the exact Notion page, Google Doc paragraph, or Slack message where the fact originated.

Browser Extension

The Dust browser extension allows users to query their assistants from any web page, enabling team members to pull company policies or customer context into web applications without switching tabs.

Enterprise Governance, Security, and Knowledge Hygiene

To maintain high response accuracy and protect company privacy over time, organizations should implement ongoing operational governance.

Combating Data Swamp Syndrome

An AI assistant's answers are only as reliable as the underlying knowledge sources:

• Prune Stale Documents: Periodically audit connected Google Drive folders and Notion pages. Archive deprecated policy documents, old product specs, and outdated procedural guides to prevent assistants from citing obsolete information.
• Curate Canonical Sources: Maintain single sources of truth for core policies. If vacation guidelines exist in both a 2022 PDF and a 2026 Notion page, remove or disconnect the older document.
• Selective Slack Indexing: Do not connect general chat channels where unverified opinions or speculative discussions dominate. Index only channels where official decisions and verified resolutions are posted.

Data Privacy and Compliance Standards

Enterprise data connected to Dust remains isolated and protected under strict security controls:

• Zero Model Training: Dust contracts with model providers (such as Anthropic, OpenAI, and Mistral) ensure that enterprise data and customer prompts are never used to train or fine-tune public foundation models.
• Access Auditing: Enterprise administrators have access to workspace audit logs showing which users queried which assistants and what data sources were referenced.
• Enterprise Authentication: Organizations can enforce SAML-based Single Sign-On (SSO) and SCIM provisioning through identity providers like Okta or Google Workspace, ensuring that employee departures immediately revoke access across all assistants and Spaces.