Dust Release Notes

Choose the model from the composer

Watch the model demo

You can now choose which model runs your agent directly from the message composer. Click the model icon to choose a specific model or one of Dust's three auto models: Basic, Standard, and Premium.

Each auto model is an ordered list of model and reasoning-effort combinations maintained by Dust. We periodically benchmark the candidates and update the lists when the preferred model changes. Dust does not route each request to a different model based on the task. When you send a message, Dust uses the first candidate on the selected list that is available to your workspace. If a candidate is unavailable, Dust continues to the next one.

The model picker works with any agent, including the default @dust agent. A selection in the composer applies to your next message in that conversation only. To make a persistent choice, set the model in Agent Builder.

The global model agents no longer appear in your agent list, but the models that powered them remain available through More models in the picker when your plan and administrator settings allow them. These agents were wrappers around a model with access to internet search, rather than full agents with their own instructions, tools, or company knowledge. …

Restrict access to model tiers

Admins can now cap members at Basic, Standard, or Premium to control which model and reasoning-effort combinations they can select. The setting applies to every model option in the selected tier, not only to the auto-model rows. Models and reasoning efforts above a member's ceiling are locked in the picker, and agents configured with an unpermitted option show an error instead of running.

Open the information panel beside the model-tier setting to see the current model and reasoning-effort options in each tier. The list changes as Dust updates its model catalog, pricing, and availability.

Available in every new credit-based workspace.

Slack MCP Improvements: Reactions, Status Updates, and Optional "Sent by Agent" Footer

🎯 What is it?

Your agents can now do more when working through the Slack integration. They can add and remove emoji reactions, update Slack status, and, when an admin allows it, send messages without the "sent by agent" footer that normally appears with personal authentication.

💡 Why is it useful?

Until now, these were hard limitations of the Slack integration: agents couldn't react to messages, change status, or send messages that looked fully native. These additions let your agents interact in Slack in a more natural way, and give you finer control over how agent-sent messages appear to your teams.

⚙ How does it work?

Reactions and status updates are available to everyone out of the box. The "sent by agent" footer can be turned off on a per-message basis, but only if an admin has enabled that option for your workspace, so you stay in control of transparency.

✨ Concrete Use Cases

Here's how you could use it:

• Acknowledging messages: An agent monitoring a support channel can add a ✅ reaction once it has picked up and handled a request, giving your team an instant visual cue.
• Signaling availability: An agent can update a user's Slack status (for example, "In a meeting" or "Focusing") based on calendar events or workflow triggers. …

GPT-5.6 Sol, Terra & Luna Are Now Available on Dust

🎯 What is it?

We've added OpenAI's newest GPT-5.6 model family to Dust. It comes in three tiers, each tuned for a different balance of speed, cost, and capability: Sol (the most capable), Terra (the balanced middle-ground), and Luna (the fastest). All three can be selected directly when building or editing an agent.

💡 Why is it useful?

Not every task needs the most powerful model, and not every task can afford to wait. Until now, choosing a model often meant trading off speed against depth of reasoning. With three clearly defined tiers, you can now match the right model to the right job, getting fast responses where you need throughput and deeper reasoning where the task demands it, without paying for more than you need.

⚙️ How does it work?

Each tier is designed for a specific profile:

• Sol: the most capable, built for complex, long-horizon agentic workflows.
• Terra: strong reasoning at a reasonable cost, a solid default for most use cases.
• Luna: optimized for high-throughput, low-latency tasks where speed matters most.

📈 Benefits for you

More control and flexibility: you can optimize each agent for speed, cost, or depth depending on its purpose, improving both efficiency and results across your workspace.

🚀 How to access it? …

🔔 Sound Notification When an Agent Needs Your Approval

🎯 What is it?

Dust can now play a sound alert whenever one of your agents pauses and waits for you to approve or decline a tool action. The sound only plays when you're not actively viewing that particular conversation, so you're notified precisely when your attention is needed elsewhere.

💡 Why is it useful?

When an agent runs a task that requires manual approval (for example, before it performs a sensitive action), it pauses until you respond. If you've switched to another tab or you're working on something else, it was easy to miss these moments and leave an agent waiting. This audio alert ensures you can step away or multitask without losing track of pending approvals.

⚙️ How does it work?

Once enabled, Dust plays your configured sound whenever an agent is waiting for approval or decline of a tool, and only if you're not currently looking at that conversation. Sound notifications are off by default, so you're fully in control of whether you use them.

✨ Concrete Use Cases

Here's how you could use it:

• Multitasking across tabs: You launch an agent on a long task, switch to your inbox, and get an audible cue the moment it needs your go-ahead, no need to keep checking back.
• Running several agents at once: While monitoring multiple conversations, you're alerted as soon as any one of them pauses for approval, so none of them sit idle.

📈 Benefits for you …

Set Default Skills in Pods

🎯 What is it?

Pod editors can now select a set of default skills directly from their Pod settings. Once configured, these skills automatically appear in the input bar of every new conversation started in that Pod, exactly as if each user had added them manually via the "/" command or the dropdown menu.

💡 Why is it useful?

Until now, if you wanted to use skills that aren't attached to a specific agent, you had to add them by hand every time you started a new conversation. This created repetitive friction, especially for teams relying on the same set of skills day to day. Default skills remove that step, ensuring the right capabilities are always ready from the very first message.

⚙ How does it work?

A Pod editor opens the Pod settings and selects the skills they want available by default. From that point on, every new conversation in the Pod automatically loads those skills into the input bar. Users can still add or remove skills per conversation as needed.

✨ Concrete Use Cases

Here's how you could use it:

• Standardizing team workflows: A Pod dedicated to customer support can have your knowledge-search and ticket-handling skills preloaded, so every agent starts with the same toolkit.
• User onboarding and activation: Set helpful skills as defaults to guide new members and help them get value from the Pod immediately, without needing to know which skills to add.

📈 Benefits for you …