Teach once.
Run again.
Browser workflows with a complete audit trail.
Show an agent a task on a website by chatting with it while a live browser does each step. Pixel Foxx turns the steps into a workflow your software calls through an API, and a person approves each run until you switch it to Auto.
Three steps from a chat to an API call
- 01
Teach
Open the website in a live browser and tell the agent what to do. Mark values that change on every run as {{name: example}}.
Fill “Amount” with{{amount: 1,240.00}}
- 02
Automate
Say “I'm done”. The agent builds a workflow with steps, variables, rules and a failure policy. Review it and confirm.
5 steps · 2 variablesMode: Review first
- 03
Call the API
Your software sends POST /v1/workflows/{id}/runs with an API key and the values. Each run waits in the approval queue unless the workflow is in Auto mode.
POST /v1/workflows/{id}/runs{ "amount": "1,240.00" }
What it does
Approval queue
See the values sent and every step the run will take, edit a value, then approve or reject.
Step repair
When a page changes, AI repairs only the step that failed. A save button that moved is held until a person confirms it.
Two-step sign-in
Authenticator codes are generated during the run. For a text-message code, the run pauses until someone enters it.
Spaces and roles
Each space keeps its own workflows, logins, keys and members. Owners, editors, operators and viewers get different rights.
Audit trail
Every filled field is read back and recorded, with screenshots before and after each step.
Webhooks
Signed events tell your app when a run needs review, needs a code, finishes or fails.
Saved logins never reach the AI model
- Saved passwords and authenticator setup keys are encrypted with AES-256-GCM before they are stored.
- The browser types them into the website. No AI model sees them.
- Row-level security in the database keeps each space separate.
- A step that saves data is never repeated automatically after it ran.