Dust agents combine workspace tool access with conversational tasks and scheduled or event-driven triggers. Giving an agent the ability to run at a later time changes the operational model: access decisions made at design time must be rechecked against permissions when the trigger actually fires.
How it works
Dust announced in April 2026 that users could create triggers for agents they could talk to, not only agents they authored. Its January 2026 documentation also described automatic tool discovery. Neither feature means the runner may ignore underlying workspace policy. A trigger’s creator, delegated service identity and agent’s effective integration permissions need independent records, because a permission revoked after scheduling should constrain subsequent jobs.
A practical, bounded workflow
For a weekly internal document summary, use a trigger referencing a specific authorized collection, store the access scope and selected agent version, then retrieve only documents visible to the initiating role. Generate a draft digest and include source links; never email external recipients without a separate approval. If the workspace reorganizes a collection before the next run, the workflow should refuse to fall back to global search.
Failure modes and limits
An agent that auto-discovers tools is powerful but creates an audit challenge: the set of tools available at runtime may change after a connector installation. Snapshot allowed capabilities per trigger, forbid new write tools by default, and alert owners on permission changes. Do not take a marketing description of trigger creation as a retention or data-residency promise.
Deployment controls
Regardless of the vendor interface, define a deployment policy for tool scopes, input trust, required approvals, call limits, credential ownership and audit retention. Grant read operations separately from externally visible writes; require a deterministic controller to validate tool arguments. A natural-language system prompt is useful guidance, not the security boundary.
When to choose this approach
Dust is useful for teams that want agents anchored to a managed workspace knowledge corpus. For fixed summaries on a static schedule, a deterministic retrieval-and-template job can be easier to review. Plan boundaries, geographical deployment and administrator capabilities need a separate verified procurement check.
What to evaluate before production
Revoke a user, disconnect an integration, relocate a document and add a newly installed tool between two trigger executions. Confirm the later run follows current authorization, creates an auditable source list and cannot turn a read-only summary into an unsupervised write action.
Version, market and tier caution (25 October 2026): This is a technical and operational guide, not a live entitlement, pricing or geographic-availability certification. Numeric commercial limits and undocumented release dates are intentionally omitted. Reconfirm the exact account, deployment region, access policy and vendor release before procurement or publication.