Multi-agent workflows

What you will do

Use multiple agents in a single conversation, set up automated triggers, and learn when to chain agents versus when to use a single agent with multiple tools.

Call multiple agents in one conversation

In Dust, you can mention more than one agent in the same message. Each agent processes the query independently and returns its own answer.

@gpt @claude Explain the pros and cons of usage-based pricing for a 20-person SaaS team.

This is useful for benchmarking: you see how different models approach the same question.

Chain agents across a conversation

You can also chain agents sequentially in a single conversation. Because all agents in a conversation share the same context, Agent B can build on what Agent A produced.

@accountSummary Give me a summary of Acme Corp.

(Agent responds with the summary)

@emailFollowup Draft a follow-up email based on the summary above.

The email agent sees the account summary in the conversation history and uses it to write a relevant email.

Use triggers for automation

Dust supports automated triggers that call an agent without you typing a message.

• Slack triggers. Link an agent to a Slack channel. Every time someone posts in that channel, the agent processes the message. Useful for auto-triaging support requests or answering FAQs.
• Scheduled triggers. Set an agent to run at a specific time (daily, weekly) with a predefined prompt.

To set up a Slack trigger, go to your agent's settings, find Slack preferences, and click Manage channels. Select the channel where the agent should listen.

When to chain versus when to use one agent

Use a single agent with multiple tools when the task is a single coherent job (research a client, then summarise, then format). Use multiple agents when the tasks require different specialisations, different data access, or different models.

Rule of thumb: if you find yourself writing instructions that say "first do X, then do Y, then do Z" and X, Y, Z require different data sources, split them into separate agents.

Test it

Set up a two-agent chain in a conversation. Verify that the second agent correctly uses the output of the first. If you set up a Slack trigger, post a test message in the channel and check the agent's response.

What you should see

Agents working together across a conversation, with each one contributing its speciality. If you set up a trigger, messages in the Slack channel should get automatic responses from the agent.