Skip to main content
Tools are the actions an agent can take inside your workspace. Every tool is enabled by default. Disabling tools you don’t need makes the agent faster, cheaper, and more accurate—the fewer options an agent has, the less likely it is to pick the wrong one.

Why scope tools

Agents choose which tool to call based on the instructions and the current state of the run. When five tools could plausibly satisfy a step, the agent has to reason about which one fits. When only one tool applies, there’s nothing to get wrong. The pattern to follow: enable exactly what the agent needs, disable everything else. An agent that writes a weekly summary doc doesn’t need channel tools, workspace tools, or the ability to create other agents. Turning those off doesn’t reduce what the agent can do—it improves how reliably it does the thing you actually want.

Projects

Actions for working with research projects, the data inside them, and the tags applied to that data.
ActionWhat it does
Create projectSpins up a new project with a name
Rename projectUpdates a project’s name
Delete projectPermanently removes a project and its contents
Add data to projectAttaches a data point (transcript, note, recording) to a project
Create highlightMarks a specific span of a transcript or doc as significant
Attach audio or videoAdds media to a project
Add utterance to transcriptAppends a new utterance to an existing transcript
Create tagCreates a new tag in the project, or applies directed existing tags
Example scoping An agent that is instructed to highlight key words or sentiment does not need Create tag tool toggled. Only Create highlight needs to be configured on.

Docs

Actions for creating and editing docs (which double as insights in Dovetail).
ActionWhat it does
Create docWrites a new doc in a specified folder
Update docEdits an existing doc
Delete docPermanently removes a doc
Add commentPosts a comment on a doc
Resolve commentMarks a comment as resolved
Edit doc contentEdits and overwrites doc content
Example scoping A weekly digest agent needs Create doc only. Turn off update, delete, and comment tools so it can’t accidentally overwrite last week’s digest.

Channels

Actions for working with Channels—the always-on ingestion layer.
ActionWhat it does
Create channelSets up a new channel
Delete channelRemoves a channel and all its data
Create topicAdds a new topic within a channel
Update topicEdits an existing topic
Add data pointIngests a new item into a channel
Side effects to know Deleting a channel removes every data point inside it. Deleted channels and their contents are recoverable—contact support if you need to restore one. Example scoping A support triage agent that reads from a channel and adds tags doesn’t need any Channel tools enabled—reading is implicit, and it isn’t creating or modifying the channel itself.

Workspace

Actions for working with folders, contacts, and custom fields.
ActionWhat it does
Create folderMakes a new folder
Rename folderUpdates a folder name
Delete folderRemoves an empty folder
Create contactAdds a new contact record
Update contactEdits contact details
Delete contactRemoves a contact
Create fieldAdds a custom field to the workspace
Update fieldEdits a field
Delete fieldRemoves a field
Side effects to know Folders can only be deleted when empty. Move or delete their contents first. Deleting a custom field removes that field’s values from every record that had it—this action is not recoverable.

Agents

A single meta-capability: create new agents.
ActionWhat it does
Create agentConfigures a new agent programmatically during a run

Combining tools

Some agents need tools from more than one category. A few common combinations:
  • Digest agent—Docs: Create doc. Everything else off.
  • Tagger—Projects: apply existing tags only. Everything else off.
  • Interview intake—Projects: Add data, Attach audio or video. Workspace: Create contact. Everything else off.
  • Triage and route—Projects: Create highlight, apply tags. Docs: Add comment. Everything else off.
Start narrow. Add a tool only when the agent tells you it needs one—if a run fails because a required action isn’t available, you’ll see it immediately and can enable exactly what’s missing.