When building an Agent in Coworker, you control exactly which tools it can access — and what it's allowed to do with them. This article explains how to add a data source to your Agent, enable write access, and require your approval before any write action is executed.
Prerequisite
The data source you want to add must be connected to your Coworker account. If it isn't, you'll be prompted to connect it during setup. See Connecting your data sources for instructions.
Adding a data source to your Agent
Open Agent Builder and open an existing Agent or create a new one.
Scroll to the Data Sources section of the Agent form.
Click Add data sources to open the searchable connector list.
Select the connector you want the Agent to use. It will appear as a row in the Data Sources section, showing the connector name, connection status, and available permission toggles.
ℹ Only grant access to the tools your Agent actually needs. Limiting data source access reduces risk and keeps your Agent focused on its task.
Enabling write access
By default, adding a data source gives the Agent read-only access — it can search and retrieve information, but cannot create, update, or send anything. To allow your Agent to take action in a tool (e.g., post to Slack, create a Jira ticket, or send an email), you need to enable write access.
For each data source that supports write actions, you'll see a Can write toggle (or Can send for messaging tools like Slack) next to the connector row.
In the Data Sources section, find the connector you want to enable write access for.
Toggle on Can write (or Can send for Slack).
For Slack, an additional channel picker will appear — select the specific channel or DM you want the Agent to send to.
Note: Turning off the "Can write" toggle switches the source back to read-only and removes any pending write configuration for that connector.
Requiring approval before write actions execute
If you want to review what your Agent intends to do before it acts, enable the Require approval option. When this is on, any write action the Agent would take is held in the Action Queue for you to review and approve first — nothing is sent or created until you give the go-ahead.
In the Data Sources section, toggle on Can write for the relevant connector.
Check the Require approval checkbox that appears below the write toggle.
Save your Agent.
When your Agent runs and a write action is triggered, it will appear in the Action Queue with an Open status. From there, you can:
Approve — the action executes immediately
Dismiss — the action is skipped without executing
To access the Action Queue, click the Action Queue button in the top bar of Agent Builder, or go to Execution History → Action Queue.
ℹ Require approval is especially useful for Agents that write to shared tools like Slack channels or your CRM, where unreviewed actions could affect other team members.
Read vs. write — what's the difference?
Access level | What the Agent can do |
Read only (default) | Search, retrieve, and read data from the source |
Can write | Create, update, send, or modify records in the source |
Can write + Require approval | Stages write actions for your review in the Action Queue before executing |
Tips
If a data source doesn't appear in the list, it may not be connected to your account — connect it first in your Data Sources settings.
If the Can write toggle doesn't appear after adding a source, try removing the source and re-adding it.
For Agents that write to shared tools, always use Require approval until you've verified the Agent behaves as expected.
Write access is scoped per Agent — enabling write for one Agent does not affect any other Agent's permissions.