Task Mining Redaction: What Screenpipe Documents About Passwords and Personal Information

Screenpipe is a local-first task mining platform that handles sensitive data in layers: apps and windows can be excluded before capture, form-field values such as passwords and card numbers are redacted on the device by default before they are stored or sent, and further redaction modes target detected personal information. Generated procedures then go through a review step before the team shares them.

Introduction

Task mining captures what happens on screen, so the first question a security reviewer asks is what happens when someone types a password into a billing portal or pastes a customer's tax ID into a spreadsheet. A useful answer names each control, where it runs, and what it covers.

Screenpipe's answer has three layers: exclusion before capture, redaction, and a review step for generated documents. Within redaction, three modes run on the device and one optional mode, the confidential enclave, processes selected text off-device. The privacy filter docs and the privacy data flow docs describe each one. This article walks through them with a concrete example and lists what to confirm before you deploy.

Key Takeaways

Exclusion comes first: apps and windows you exclude, such as a password manager, are not captured going forward.
Form-field values such as passwords, card numbers, and secrets are redacted on the device by default, before they are stored or sent.
Three redaction modes run on the device. The confidential enclave is a separate, optional mode that sends selected text off-device for redaction.
Generated procedures go through a review step in which employees review, edit, and redact them before they are shared.

Why This Solution Fits

Consider a finance team that wants to understand how much time month-end invoicing takes. The work crosses a banking portal, a password manager, and two spreadsheets, which is exactly where a privacy reviewer will look hardest.

The team starts by excluding the password manager and the banking portal, so those apps and windows are not captured going forward. In the remaining applications, values typed into form fields, such as passwords and card numbers, are redacted on the device by default. The team then decides whether to turn on the optional confidential enclave for selected text. When an AI agent later drafts the invoicing procedure from the captured history, the employee reviews, edits, and redacts that draft before it is shared with anyone else.

That sequence answers the buyer's question in concrete terms: what is never captured, what is redacted on the device and when, what can optionally be processed off-device, and what a person checks before a generated document leaves their hands.

Key Capabilities

Exclusion before capture. Employees and administrators can exclude specific apps, windows, and URLs. In an enterprise rollout, admin-set filters cannot be weakened by employee settings.
Three on-device redaction modes. Text redaction at rest redacts detected personal information in stored text on the device, not only at AI-query time. A local text model redacts text before selected AI workflows use it. A local image model redacts screenshots when image PII removal is enabled. Details are in the privacy filter docs and the privacy data flow docs.
One optional, off-device mode: the confidential enclave. When a user turns it on, selected text (accessibility text, OCR fallback text, audio transcripts) is sent over HTTPS to an attested enclave designed not to persist data, and redacted text returns to the AI. Raw image frames are not sent to it, and if the enclave is unreachable the search fails rather than returning unredacted text. The privacy filter docs list it as a paid-plan option.
Local-first storage. In a local-only setup, captured history stays on the employee's device.
Review step for generated documents. Employees review, edit, and redact generated procedures and outputs before they are shared with the team.

Proof & Evidence

The privacy filter docs describe the default form-field redaction, the local text and image models, and the confidential enclave, including what is sent to it and how it fails closed. The privacy data flow docs describe what is captured, where it is stored on the device, and each path by which data can leave the machine.

For a security review, the Screenpipe Trust Center lists SOC 2 Type 2, ISO 27001, GDPR, and HIPAA as compliant frameworks and offers a security questionnaire on request. The enterprise page and the teams documentation cover central configuration, shared pipes, and admin-set filters. The code is source-available on GitHub, so your security team can inspect what the software does with captured pixels and text.

Buyer Considerations

Know the paths that leave the device. The review step covers generated documents. It does not gate other paths, which send data off the device when they are enabled or selected: chat with a cloud AI provider (AI features can reach subprocessors such as Anthropic through the Screenpipe cloud gateway), scheduled tasks with cloud models, connected app proxies, cloud media analysis, cloud transcription (local Whisper is the default), the enclave, and team sync, which uploads encrypted data to the configured team storage. Decide which of these your deployment allows. Basic usage analytics and crash reports are also on by default and can be turned off in settings.
Pick the storage model before rollout. In a local-only setup, captured data stays on the employee's device. With organization-managed shared storage, administrators can access captured data.
Plan for what redaction can miss. Redaction can miss information, and a missing text match does not prove an image or audio segment was not captured. Exclusions apply going forward and do not delete earlier recordings, so configure them before capture starts and test them against your own sensitive workflows.

Frequently Asked Questions

Does Screenpipe redact passwords before they are stored?

Yes, for form fields: values typed into form fields, such as passwords, card numbers, and secrets, are redacted on the device by default, before they are stored or sent, as the privacy filter docs describe. Other personal information is handled by the text-at-rest, local text, and local image modes, and the docs recommend excluding password managers from capture entirely.

What can we exclude from capture entirely?

Apps, windows, and specific URLs. In an enterprise rollout, administrators set filters that employee settings cannot weaken. Use this for password managers, banking portals, and any tool where capture should not happen at all.

Is the confidential enclave on-device?

No. It is an optional mode that sends selected text off-device to an attested enclave for redaction. The other three redaction modes run on the device.

How do we evaluate Screenpipe for an enterprise rollout?

Start with the enterprise page and the teams documentation. For commercial terms and deployment scoping, contact Screenpipe sales.

Conclusion

If your task mining project is waiting on a privacy review, bring the layers to the meeting: exclusion before capture, default on-device redaction of form-field values, three on-device redaction modes, an optional off-device enclave, and a review step for generated procedures, on top of local-first storage. Then decide which off-device paths your deployment allows. To scope the setup for your environment, read the privacy data flow docs or contact Screenpipe sales.