Computer Agents

AI agents that operate websites for you, for systems that have no API.

Feathery Computer Agents operate websites the way a person would. An agent opens a browser, reads the page, logs in, fills out forms, submits them, and downloads files. Use an agent for systems that have no API, such as carrier portals, legacy admin consoles, and government filing sites.

Computer Agents are available on the Business plan. Reach out to sales to get access.

Create a Computer Agent

Go to the Computer Use tab of your account and click Create Computer Agent. You need the Admin or Editor role.

1. Give your agent a name.

2. Describe what it should do. This is the agent's task.

3. Choose a schedule, or leave it on Manual.

Write the Task

The task is the agent's complete set of instructions, written in plain English. The agent has no other knowledge of your workflow.

Go to https://portal.example.com and log in with PORTAL_USERNAME and PORTAL_PASSWORD.
Open the Statements tab, find the most recent monthly statement, and download it.
Report the statement date in your notes.

• Name any credential you want the agent to use by its key.

• Reference form fields with {{field_id}} syntax. See Pass Form Data into the Task.

• Name the landmarks the agent should look for, such as "the Statements tab", rather than scripting each click.

Agent Settings

• Properties: the agent's name, its ID, the task, the file field that receives downloads, and the agent's autonomy — its run mode and approval window.

• Schedule: how often the agent runs on its own.

• Browser Policy: the page each run opens to, and the domains the agent can reach.

• Credentials: logins the agent needs on the site.

Run Modes

The run mode sets how much the agent may do without you. Agents created from the dashboard start supervised.

• Supervised: the run pauses for your approval before risky actions like submitting forms, making payments, or entering credentials. The agent can also pause to ask you questions when the task is missing information only you can supply.

• Autonomous: runs start to finish without pausing for approval. Nobody is watching an autonomous run, so it refuses to move money outright.

• Observe: a dry run. The agent navigates and reads the site but takes no actions, then reports exactly what it would have done, step by step.

A supervised run that pauses holds its place and waits for the approval window — from 5 minutes to 72 hours, set alongside the run mode. If the window closes without a decision, the run fails. See Review a Run for how to respond to a paused run.

Next Steps

Credentials & Browser PolicyTrigger a RunReview a Run