Dust Release Notes

What is it?

Gemini 3.5 Flash, Google's latest AI model, is now available on Dust. It is designed for speed and delivers frontier-level performance across a wide range of tasks — making it one of the fastest capable models available on the platform.

Why is it useful?

When working with agents that handle high volumes of requests, or when you need quick turnaround on complex tasks, model speed matters. Gemini 3.5 Flash combines the responsiveness you'd expect from a fast model with the quality benchmarks of a frontier one — a combination that's rarely available in one package.

How does it work?

Gemini 3.5 Flash can be selected as the underlying model when configuring any agent on Dust, just like other available models. It integrates seamlessly into your existing workflows with no additional setup required.

Concrete Use Cases

Here's how you could use it:

• High-volume support agents: Handle large numbers of customer queries faster, without sacrificing response quality.
• Real-time drafting agents: Speed up content generation workflows where low latency makes a meaningful difference in team productivity.

Benefits for you

Faster responses, frontier-level quality, and no trade-offs on accuracy. Gemini 3.5 Flash gives your agents more horsepower for time-sensitive or high-throughput use cases.

How to access it?

Gemini 3.5 Flash is available to all users today. To use it, open any agent's settings, navigate to the model selection section, and choose Gemini 3.5 Flash from the list.

LLM model refresh: Grok 4.3 upgrade + Anthropic model migrations

📌 Context

Two of our LLM providers are making changes to their model catalogs:

• xAI has upgraded Grok 4 to version 4.3 on their infrastructure
• Anthropic is retiring several Claude models on June 15, 2026, and has already retired the Haiku 3.5 models

To ensure continuity of service and avoid any disruption, we're proactively updating Dust's model catalog.

🔄 Impact on Dust

We've already completed the following updates:

Grok models (live now):

• Our "grok-4" models now automatically run on Grok 4.3

Anthropic migrations (rolling out before June 15):

• claude-4-sonnet → migrating to claude-sonnet-4-6
• claude-4-opus → migrating to claude-opus-4-7
• claude-3-haiku / claude-3-5-haiku → migrating to claude-haiku-4-5
• claude-sonnet-4-5 → migrating to claude-sonnet-4-6

👤 Impact for you

No disruption expected. Your agents and workflows will continue to function normally:

• If you're using Grok models, you're already benefiting from the 4.3 upgrade
• If you're using the affected Anthropic models, we'll handle the migration seamlessly before the retirement dates
• Your agent configurations don't need to be updated—we're managing the transition on the backend

✅ Actions required

No action required on your part. We're handling these updates automatically to ensure uninterrupted service.

If you have any questions about these model changes or want to understand the differences between the new model versions, feel free to reach out to your Customer Success Manager.

What is it?

You can now create a "branch" from any existing conversation. This creates a new conversation that inherits key context from the original—like a summary of what happened, attachments, and tool outputs—without carrying over the full message history.

Why is it useful?

As conversations evolve, you often reach a point where you need to split work into parallel streams or share results without exposing all the back-and-forth. Branching lets you divide complex projects into focused workstreams, or collaborate on outcomes while keeping the messy details contained.

How does it work?

Access the conversation menu (top right of the screen or next to the conversation title in the left sidebar) and select the branching option. The new conversation is created in 10-20 seconds with a compacted summary of the original conversation, plus copies of any attachments and tool outputs. The branch inherits the same access permissions as the original conversation at the time of creation.

Concrete Use Cases

Here's how you could use it:

Split a project into parallel workstreams: Your conversation outlines a plan for building a new feature with three components. Branch into three separate conversations—one for each technical design—so teams can work independently without cluttering a single thread.

Share results, not process: You've built a detailed data frame after hours of iteration and refinement. Branch the conversation so collaborators can access the final output and continue refining it, without wading through all the trial and error.

Benefits for you

• Better organization: Keep complex projects manageable by splitting them into focused conversations
• Cleaner collaboration: Share outcomes and context without overwhelming teammates with unnecessary details
• Flexible workflows: Move seamlessly from planning to execution across multiple parallel efforts

How to access it?

Look for the conversation menu in the top right corner of your screen, or next to the conversation title in your left sidebar. The branching option is available to everyone—try it on your next multi-part project!

Asana MCP Migration to V2 – Action Required

📌 Context

Asana has shut down their MCP (Model Context Protocol) V1 endpoint. To ensure continued integration between Dust and Asana, we've migrated all existing Asana MCP servers to V2, which uses a more secure static OAuth authentication method.

🔄 Impact on Dust

We've proactively migrated all Asana MCP server configurations to the new V2 endpoint to minimize disruption. The infrastructure is ready and working.

👤 Impact for you

If you use Asana MCP in your workspace: Your existing Asana connection has been reset and needs to be reconfigured. Until you complete the setup steps below, your agents won't be able to access Asana data.

If you don't use Asana MCP: No impact – you can ignore this update.

✅ Actions required

For workspace admins using Asana MCP:

Step 1: Create an app on Asana

1. Go to Asana My Apps and click Create new MCP app
2. In the OAuth tab, add this redirect URL: https://dust.tt/oauth/mcp_static/finalize
3. Note your Client ID and Client Secret (you'll need these in Step 2)

Step 2: Configure in Dust

1. Navigate to Spaces > Tools in your Dust workspace
2. Find your Asana MCP server (it will show as disconnected)
3. Fill in the following details:
   • Client ID and Client Secret from your Asana app
   • Authorization Endpoint: https://app.asana.com/-/oauth_authorize
   • Token Endpoint: https://app.asana.com/-/oauth_token
4. Click Setup connection and complete the authorization flow

📚 Need more details? Check out the Asana MCP integration documentation.

What is it?

Enterprise workspace administrators now have access to comprehensive audit logs directly in the admin panel. These logs track all agent executions, tool calls, and administrative actions across your workspace. Additionally, you can stream these logs in real-time to your existing Security Information and Event Management (SIEM) platforms like Datadog, Splunk, or any custom HTTPS endpoint.

Why is it useful?

Security and compliance teams need complete visibility into workspace activities for several critical purposes: responding to security incidents, meeting regulatory compliance requirements, conducting internal audits, and distinguishing between actions taken by users versus those performed by agents. Until now, this level of detailed activity tracking wasn't available in Dust.

How does it work?

Audit logs are accessible through your admin panel with real-time updates. You can view the complete history of activities, filter by type (agent executions, tool calls, admin actions), and configure streaming to your preferred SIEM platform or custom endpoint for centralized security monitoring.

Concrete Use Cases

Here's how you could use it:

Security Incident Investigation: When investigating a potential security event, quickly review which agents were executed, what tools they used, and which users or agents performed specific actions during the relevant timeframe.

Compliance Reporting: Generate audit reports for compliance certifications (SOC 2, ISO 27001, GDPR) by exporting complete activity logs showing data access patterns, administrative changes, and agent behaviors.

Separation of Concerns: Clearly distinguish between actions initiated by human users and those performed autonomously by agents, helping you understand and control automated workflows in your environment.

Benefits for you

• Enhanced Security Posture: Complete visibility into all workspace activities enables faster threat detection and incident response
• Simplified Compliance: Meet regulatory requirements with comprehensive, exportable audit trails
• Operational Intelligence: Understand how agents and users interact with your Dust workspace to optimize workflows
• Centralized Monitoring: Integrate with your existing security infrastructure for unified monitoring across all enterprise tools

How to access it?

This feature is available exclusively for Enterprise plan workspaces. Workspace administrators can access audit logs by navigating to the admin panel. For detailed setup instructions, including SIEM streaming configuration, visit our documentation at

What is it?

We've shipped a new file system for conversations. The first thing it unlocks: agents can now see images returned by MCP tools. When a tool produces an image (JPEG, PNG, GIF, WebP), the agent receives the actual visual content and can interpret, analyze, and describe it.

Why is it useful?

Until now, when an MCP tool returned an image, agents were blind to it. They could acknowledge the file existed but couldn't look at it. This was a real gap: screenshots from Microsoft Drive, charts from data tools, visual outputs from external integrations all invisible to the agent. That's now fixed.

How does it work?

Behind the scenes, all conversation files are now organized in a unified, structured system. When an agent needs to access an image returned by an MCP tool, it receives the actual visual content instead of just metadata.

Concrete Use Cases

Here's how you could use it:

• Screenshot analysis: Have an agent interpret a screenshot returned by a Drive or browser tool without manually describing what's on screen.
• Chart and graph reading: Ask an agent to extract trends or key metrics from a chart generated by a data tool.
• Visual tool output review: Any MCP tool that produces images can now have its output directly analyzed by the agent in context.

What is it?

Context Compaction is a new feature that automatically summarizes earlier parts of long conversations with your agents. Instead of losing context or hitting limits, Dust intelligently condenses the conversation history while preserving the essential information your agent needs to continue working effectively.

Why is it useful?

During extended work sessions—like analyzing complex documents, iterating on code, or working through multi-step projects—conversations can become very long. Previously, this could degrade agent performance or force you to start a new conversation and lose your working context. Context Compaction solves this by maintaining quality throughout long interactions while keeping your full conversational context available in summarized form.

How does it work?

A new usage indicator appears in your input bar showing how much context is being used. The system guides you through three stages: at 30% usage, compaction becomes available; at 70%, you'll see a helpful reminder to compact; and at 80%, the system will pause new messages until you run compaction to ensure optimal performance.

Concrete Use Cases

Here's how you could use it:

Complex Analysis Projects: You're working with an agent to analyze multiple quarterly reports over several hours. Instead of splitting your work across multiple conversations, Context Compaction lets you maintain the full analytical thread while summarizing earlier findings.

Iterative Development: You're collaborating with an agent on code development with many iterations and refinements. The conversation naturally grows long, but compaction preserves all your decisions and context without performance degradation.

Benefits for you

Save time by never needing to restart conversations or re-explain context. Maintain better continuity in complex projects. Work confidently knowing your agent has access to the full conversation history in an optimized format.

How to access it?

Context Compaction is now available to all users. Simply watch for the context usage indicator in your input bar—when it appears, you'll be guided through the process automatically. Learn more in our documentation.

What is it?

Your agents can now apply labels to emails and archive messages directly in Gmail. These new capabilities extend the existing Gmail tools, giving your agents more control over inbox management.

Why is it useful?

Many workflows require not just reading emails, but organizing them too. Whether you're triaging customer requests, categorizing vendor communications, or cleaning up your inbox, agents can now take action automatically—no manual sorting required. Combined with Wake-ups (scheduled agent runs), you can fully automate inbox maintenance.

How does it work?

Agents with access to the Gmail tools can now use new actions to apply any of your existing Gmail labels and move emails to archive. These actions work alongside the existing Gmail capabilities your agents already have.

Concrete Use Cases

Here's how you could use it:

• Automated inbox triage: Set up an agent that runs every morning via Wake-ups, reviews new emails, applies labels like "Urgent", "Follow-up", or "Read Later", and archives low-priority messages.

• Customer support categorization: Have an agent automatically label incoming support emails by topic (Billing, Technical, Feature Request) and archive resolved threads, keeping your team's shared inbox organized.

• Vendor communication management: Create an agent that identifies invoices, contracts, and purchase orders in your email, applies the appropriate labels, and archives them after filing the information in your system.

Benefits for you

• Save time: Eliminate manual email sorting and filing
• Stay organized: Maintain a clean, well-labeled inbox automatically
• Enable new workflows: Combine with Wake-ups for fully autonomous inbox management
• Reduce noise: Archive processed emails so you can focus on what matters

How to access it?

The new label and archive capabilities require additional Gmail permissions. To enable them:

1. Go to Personal Settings (bottom left of Dust)
2. Find your Gmail connection and disconnect it
3. The next time an agent needs Gmail access, you'll be prompted to re-authenticate with the updated permissions

Once re-authenticated, your agents will automatically have access to the new label management and archiving tools.

What is it?

Agents can now set their own wake-up schedules to resume work at a future time within an ongoing conversation. Think of it as giving your agent the ability to "set a reminder" for itself to check back on something, run a recurring task, or wait for a response before continuing.

Why is it useful?

Sometimes work doesn't happen all at once. You might need to wait for someone to respond, check if a condition has changed, or simply run the same process every day at 9 AM. Until now, you'd have to manually come back and prompt the agent again. With wake-ups, the agent handles the timing for you automatically.

How does it work?

Agents have access to a wake-up tool that lets them schedule themselves to continue the conversation at a specific time or on a regular schedule (daily, weekly, etc.). Once scheduled, the agent will automatically "wake up" and continue where it left off.

Concrete Use Cases

Here's how you could use it:

• Follow-up automation: Ask an agent to send an email to a colleague and check back in 2 hours to see if they've responded, then proceed with next steps based on their answer.
• Recurring updates: Have an agent refresh a data dashboard every Monday at 9 AM, or check project status every Friday afternoon and send you a summary.
• Real-time monitoring: Set an agent to check an external system every 30 minutes until a specific condition is met (like a deployment completing or a document being approved).

Benefits for you

No more manual follow-ups or remembering to re-prompt your agents. You can now set up truly autonomous workflows that span hours, days, or weeks, with the agent managing its own schedule and keeping work moving forward without your intervention.

How to access it?

Wake-ups are now available to all users. Simply ask your agent to "check back in [time]" or "run this every [schedule]" and it will use the wake-up tool automatically. For more details on how to configure wake-up schedules, check out the documentation:

https://docs.dust.tt/docs/wake-ups

What is it?

You can now connect Gong to Dust through our MCP (Model Context Protocol) integration. This allows your agents to access call transcripts and notes directly during conversations and workflows, without leaving Dust.

Why is it useful?

Sales and customer success teams have valuable insights locked in Gong recordings. By connecting Gong as a live tool, your agents can pull context from customer calls on-demand, turning those conversations into actionable intelligence right when you need it—whether you're preparing for a meeting, writing follow-ups, or analyzing trends.

How does it work?

Once connected, Gong becomes available as a tool that your agents can use. When an agent needs information from a call, it queries Gong in real-time and retrieves the relevant transcript or notes to inform its response.

Concrete Use Cases

Here's how you could use it:

Pre-meeting preparation: Ask an agent to "Summarize the last 3 calls with Acme Corp" and get instant context before your next meeting.

Customer insight synthesis: Create a workflow that pulls key objections or feature requests from recent calls and compiles them into a weekly report.

Follow-up automation: Have an agent draft personalized follow-up emails that reference specific points discussed in the most recent Gong call.

Benefits for you

• Instant access: No manual searching through Gong—your agents retrieve exactly what they need
• Better context: Agents can reference actual customer conversations to provide more relevant responses
• Time savings: Eliminate copy-pasting between tools and streamline your workflow

How to access it?

This feature is available to all workspaces. Check out our documentation to set up the integration:
https://docs.dust.tt/docs/gong-mcp

Note: This is a live, tool-based integration. If you're looking to synchronize transcripts into Dust for semantic search across all your data, use the Gong connector instead.

What is it?

You can now type / directly in the conversation input bar to quickly access and add capabilities to your conversations. A searchable dropdown appears instantly, listing all available skills and MCP tools, which you can filter and select using your keyboard or mouse.

Why is it useful?

Previously, accessing capabilities required navigating through the toolbar, which could interrupt your workflow. This new slash command feature streamlines the process, letting you stay focused on the conversation while quickly adding the tools you need—similar to how modern text editors and communication tools work.

How does it work?

Simply type / in the input bar, and a dropdown menu appears with all your available capabilities. Continue typing to filter results using fuzzy matching (you don't need to type exact names), then select what you need with your keyboard (arrows, Enter, Tab) or mouse. Press Escape to close the menu.

Concrete Use Cases

Here's how you could use it:

• Quick data analysis: Type / then "calc" to instantly find and add calculator or data processing tools without breaking your thought process
• Adding specialized skills mid-conversation: Type / then start typing a skill name like "research" to quickly enable research capabilities when you realize you need them

Benefits for you

This feature saves time and keeps you in flow. Instead of moving your cursor to the toolbar, you can access everything through keyboard shortcuts, making conversations with agents faster and more efficient—especially valuable when you're working through multiple tasks quickly.

How to access it?

The feature is available to everyone right now. Just type / in any conversation input bar to try it out.

What is it?

OpenAI's latest model, GPT 5.5, is now available on Dust. You can select it when building custom agents or use it directly through the global agent interface.

Why is it useful?

GPT 5.5 represents OpenAI's newest advancement, offering improved performance over the previous GPT 5.4 model. This means better reasoning, more accurate responses, and enhanced capabilities for your agents.

Concrete Use Cases

Here's how you could use it:

• Enhanced analysis agents: Build agents that handle complex research, data analysis, or strategic planning with improved reasoning capabilities
• Improved conversational agents: Create customer-facing or internal support agents that provide more nuanced and accurate responses

Benefits for you

Access to cutting-edge AI technology means your agents can deliver higher quality outputs, handle more sophisticated tasks, and provide better assistance to your team.

How to access it?

• For custom agents: Open the agent builder and select "GPT 5.5" from the model dropdown menu
• For quick tasks: Use the global agent which now runs on GPT 5.5 by default

What is it?

OpenAI has released GPT Image 2, a new image generation model that's now the default for all image creation on Dust. This model brings significant improvements in image quality, detail rendering, and the ability to accurately generate readable text within images.

Why is it useful?

Previous image generation models often struggled with two key challenges: editing existing images effectively and incorporating clear, readable text into generated visuals. GPT Image 2 addresses both of these limitations, opening up new possibilities for creating professional-quality images that include precise text elements—something that was previously difficult or impossible to achieve consistently.

How does it work?

GPT Image 2 is automatically used whenever you generate images through Dust agents or workflows. The model excels at understanding complex prompts, maintaining high fidelity to your specifications, and rendering fine details with unprecedented precision.

Concrete Use Cases

Here's how you could use it:

Marketing Materials: Generate branded images with company names, slogans, or product descriptions clearly visible and professionally rendered—perfect for social media posts, presentations, or campaign materials.

Data Visualization Enhancement: Create infographic-style images with charts, labels, and annotations that include actual readable data points and explanations, making complex information more accessible.

Image Editing & Iteration: Take existing images and ask your agent to modify specific elements while preserving the overall composition—useful for refining visual assets or creating variations of existing designs.

Benefits for you

You can now generate more professional, production-ready images directly within your Dust workflows. The ability to include clear text eliminates the need for post-processing in external design tools, saving time and streamlining your creative processes. The improved editing capabilities also mean you can iterate on images more efficiently.

How to access it?

No action needed—GPT Image 2 is already the default image generation model on Dust. Simply continue using image generation in your agents as you normally would to automatically benefit from these improvements.

What is it?

Agents can now proactively pause and ask you structured questions mid-conversation when they need clarification. Instead of guessing or making assumptions, they'll present you with single or multi-select options to help guide their next actions.

Why is it useful?

Until now, when an agent faced ambiguity, it would either guess (sometimes incorrectly), take a random path, or get stuck trying to interpret unclear instructions. Now, agents can simply ask you directly—turning uncertainty into a quick, interactive exchange. This prevents wasted time, reduces errors, and makes conversations feel more collaborative and intuitive.

How does it work?

When an agent needs input to proceed, it will pause and present you with a clear question and predefined answer options (single-choice or multiple-choice). You select your answer, and the agent continues with exactly the information it needs.

Concrete Use Cases

Here's how you could use it:

• Research workflow: You ask an agent to "analyze our competitors." Instead of picking arbitrary companies, it asks: "Which competitors should I focus on?" with options like [Company A, Company B, Company C, All of the above].
• Content formatting: You request a report summary. The agent asks: "What format do you prefer?" with options like [Bullet points, Paragraph form, Executive summary].
• Data prioritization: You ask for insights from multiple sources. The agent clarifies: "Which data sources should I prioritize?" offering [Internal reports, Public data, Customer feedback, All sources].

Benefits for you

• More accurate outputs: Agents work with your explicit input instead of assumptions
• Faster resolution: No back-and-forth to correct misunderstandings
• Better control: You guide the agent's direction at key decision points
• Richer interactions: Conversations feel more natural and collaborative

How to access it?

This feature is available by default for all agents on Dust and Deep Dive, and works seamlessly in Slack conversations as well. No configuration needed—your agents will automatically ask questions when they need clarification.

Added

What is it?

You can now import skills directly into Dust from a GitHub repository or by uploading a .zip file from your computer. This new capability gives you more flexibility in how you manage and deploy your skills across your workspace.

Why is it useful?

If you're managing multiple skills or working with a team that maintains skills in a version control system, you previously had to manually copy and paste code into Dust. This new import feature allows you to centralize your skills in GitHub (or any other repository) and keep them in sync with your CI/CD pipeline, ensuring your Dust agents always use the latest versions.

How does it work?

Simply provide a GitHub repository URL or upload a .zip file containing your skill files. Dust will import the skill structure and make it available in your workspace.

Concrete Use Cases

Here's how you could use it:

Development teams maintaining shared skills : Your engineering team maintains a repository of company-specific skills (e.g., data formatting, internal API integrations). When you update the skill in GitHub, you can quickly re-import it to Dust to keep all agents synchronized.

Distributing skills across workspaces : You've built a powerful skill for market research and want to deploy it across multiple Dust workspaces (different departments or clients). Export it once as a .zip and import it wherever needed.

Benefits for you

• Version control : Keep your skills in Git alongside your other code, with full history and collaboration features
• Automation : Integrate skill updates into your existing CI/CD workflows
• Portability : Easily share and duplicate skills across workspaces without manual copying
• Consistency : Ensure all team members are using the same version of your custom skills

How to access it?

This feature is available to everyone. When creating or updating a skill, look for the new import options that allow you to specify a GitHub repository URL or upload a .zip file.