Product documentation

What Pacific Machines is

Pacific Machines helps a company decide what to automate with AI by learning from approved real work. It finds recurring patterns and exceptions, then turns that evidence into inspectable insights, opportunities, and reports.

A person reviews the evidence and decides what should move forward. The product does not deploy automations on its own.

Decide what to automate first, with evidence.

Pacific Machines is the public name of both the company and the desktop app. The legal entity is Reefo, Inc.

Who it is for

• Operators who know where work hurts but cannot yet prove what deserves automation.
• Operational leaders who need evidence before funding an AI project.
• Teams that want to compare work across people without starting from a blank process map.

Install and sign in

Pacific Machines is a desktop app for macOS and Windows. This website does not offer a self-serve software download.

To get the installer, book a call. We share the current signed build after that conversation.

Book a call

Supported platforms

Intel Mac is not offered.

First launch

1. Open the installer we shared with you (macOS disk image or Windows setup). On Windows and macOS the app shows as Pacific Machines (or Pacific Machines Preview on preview builds).
2. Sign in to your Pacific Machines account and organization.
3. Complete onboarding: describe your role, then grant the permissions Learning needs.
4. Start a Learning session and do normal work.

Once installed, newer builds install through the signed in-app updater. If your install is below the minimum required version, Pacific Machines shows a clear Update action on macOS and Windows.

Learning from work

Capture begins on your computer. When Learning is on and the operating system has granted the needed permissions, Pacific Machines observes approved accessibility and interaction signals while you work.

The raw record is written only on that computer, in ~/.jarbas/db.sqlite. It includes bounded accessibility text, the focused app and window, available document or repository context, focus transitions, clicks, shortcuts, and semantic typed or pasted text after local redaction. That database stays on the computer. It is not uploaded.

Learning does not capture screenshots, pixels, OCR frames, video, or images. It does not capture microphone or system audio, store accessibility trees, retain raw keystrokes, or save pointer coordinates. Meeting transcription is not part of the product.

A small Learning pill can stay on screen (near the top center) so you can pause, resume, or return to the app without hunting for the window.

Permissions the app may request

You can skip a permission during onboarding and finish later in Settings. Revoking Accessibility or Input Monitoring on macOS stops capture. Pacific Machines cannot override operating-system privacy controls. Learning does not request Screen Recording because it does not read pixels.

Using Learning

1. Turn Learning on in the app (or from the Learning pill on your desktop).
2. Do your usual work. Capture skips apps and sites you have ignored.
3. Pause or stop Learning when you do not want observation.
4. Run Update Brain (or let the built-in daily schedule do it) so captured days become Brain notes, then use Recaps, Opportunities, and Chat.

What stays local

Learning evidence is stored on your computer under ~/.jarbas by default. The capture database, including raw accessibility text, is not uploaded. Some features use the cloud on purpose: sign-in, the Daily Recap activity log, Brain notes, shared team reports, plugins you connect, and model inference when you use Chat or Agent.

For file locations, cleanup, and deletion, read the Privacy Policy.

Workspace: Chat, Brain, Recaps, Opportunities

After Learning has something to work with, the desktop app is organized around a few surfaces.

Chat

Ask questions about the work Pacific Machines has already learned. Chat can pull local capture context, Brain notes, recaps, and — if you connected them — optional plugins. When you send a message, relevant context is sent to the model provider for that request. The whole capture library is not uploaded.

Asking Chat about your day does not pause Learning; capture keeps running while Chat looks up history.

Chat can also run named skills, such as generating a recap or a report, and can write Brain facts when you ask it to remember something.

Brain

Brain is your local Markdown workspace under ~/.jarbas/brain/: identity, company context, workflows, and related notes. Browse it in Graph, Preview, and Files views. Daily Recaps are a separate activity log, described below. They are not a story inside Brain.

• Update Brain commits captured days through yesterday into Brain notes and workflows, and can sync those documents when you are signed in.
• Check Brain reports whether captured days still need Update Brain, without blocking your work.

Built-in schedule (fires only while the app is open):

You can pause or resume that schedule. Older weekly / monthly / yearly rollup schedules are retired.

Daily Recaps

A Daily Recap is a file the app writes from the local capture record, one per day, in time order. It is an activity log, not a story and not a copy of the raw record.

A stretch names the time, the app, and the window, then lines for what was opened, clicked, shortcut, pasted, or typed. Raw accessibility text is not copied into it and stays in ~/.jarbas/db.sqlite on the computer.

While recording is on, that activity log is sent to your organization a few times during the day. What is sent is the log. It is not a screenshot, pixel data, a video, an image, the database, or raw accessibility text. After the local day has ended, a short summary may be written in the cloud from the log. The summary is not written from the raw capture.

Opportunities

Find Opportunities is Brain-first: it reads what Update Brain already learned (especially workflows), instead of waiting on a separate multi-stage evidence pack or a storage-size gate.

1. Capture work with Learning.
2. Run Update Brain (and Check Brain if you want a status check).
3. Find Opportunities from the committed Brain.

Each opportunity is meant to be inspected: evidence, confidence, and a writeup a person can accept, reject, or refine. A person still decides what should move forward. Building a skill from an opportunity is an explicit step, not an automatic deployment.

Plugins

Optional connectors add context from tools you already use — for example mail, calendar, issue trackers, or documents. They enrich recaps and Chat. They never replace locally observed work.

You choose which tools to connect. Connected-app reads are explicit. Disconnect a plugin in the app when you no longer want it.

Pacific Machines may call a connected service only for tasks you approve (for example sending a message). Plugin credentials are managed through the connection flow, not pasted into a generic secrets file in Settings.

Team reports and the Team Dashboard

While recording is on, the Daily Recap activity log is sent to the organization a few times during the day so teammates with access can read the day. That upload is the log. It is not a screenshot, pixel data, a video, an image, the capture database, or raw accessibility text.

Team Dashboard

Admins can open a Team Dashboard built from members' synced Daily Recaps. It shows numbers the organization can inspect: active time, focus, context switches, apps used, and coverage. It does not read another person's capture database or accessibility trees.

Coverage is labeled complete, partial, or sparse so comparisons are not treated as precise when too few days were observed.

Team narrative recaps

A separate admin action can write an organization-level narrative from members' synced Daily Recaps. That narrative is not a second copy of anyone's raw capture.

Shared reports

Chat can generate an HTML report from synced recaps for people who should review the same evidence. Finished reports for the organization are stored in the cloud workspace so teammates with access can open them.

Local deletion of capture on one computer does not remove shared reports or synced Brain documents already stored for the organization. Those are managed from Team by someone with access.

Privacy controls

You control what is captured and what is shared.

• Pause capture by turning Learning off. On macOS, revoking Accessibility or Input Monitoring also stops capture.
• Ignore lists skip capture while an ignored app or URL is in focus. Saving the list applies going forward and does not delete past captures.
• Text cleanup finds common secrets and personal data in stored text and replaces matches with labeled placeholders. It edits stored text in the local database. It does not upload that database.
• Delete local data from Settings → Storage for a date range, or reset all local Pacific Machines data on this computer (~/.jarbas).

Sensitive text is redacted before it reaches local storage, and text is dropped if redaction fails. The separate cleanup tool is pattern-based and can miss sensitive data that does not match a rule. Older installs may still have screenshot files from an earlier capture mode until those files are deleted; current Learning never reads or writes screen pixels.

The full Privacy Policy is the technical reference for data flows, file layout, retention, and limitations.

FAQ

Does Pacific Machines record my microphone or meetings?

No. It uses accessibility and interaction signals. Audio is not captured.

Does Learning capture screenshots, video, or images?

No. Learning does not capture screenshots, pixels, OCR frames, video, or images. Bounded accessibility text and interaction data are written to ~/.jarbas/db.sqlite on your computer and stay there. Accessibility trees are not stored. While recording is on, a Daily Recap is sent a few times during the day. That file is an activity log of what was opened, clicked, shortcut, pasted, or typed. It is not an image, raw accessibility text, or the database.

Where is my Learning data stored?

On your computer under ~/.jarbas by default, including the raw capture database and its accessibility text. Sign-in, the Daily Recap activity log, Brain notes, shared team reports, plugins you connect, and Chat/Agent inference use cloud services as described in the Privacy Policy. The raw capture library is not mirrored to a remote learning store.

Does Chat pause Learning?

No. Asking Chat what you were doing keeps Learning running. Answers about your day come from capture history and Brain notes.

Can my employer see my raw Learning history?

Teammates with access can see synced Daily Recap activity logs, Brain notes, opportunities, and reports your organization chooses to share. They do not get another person's capture database or raw accessibility text. Team Dashboard numbers come from those synced logs, not from live capture.

What happens if I deny permissions?

On macOS, Learning cannot capture without Accessibility and Input Monitoring. Other features (Chat on already-learned data, Brain notes already on disk, account settings) can still be available. You can grant permissions later in system settings and in the app. Screen Recording is not required.

Does it work on Intel Macs or Linux?

No public Intel Mac channel is offered. The public beta is Apple silicon on macOS 11+ and x64 Windows 10/11.

Will it automatically deploy automations?

No. Pacific Machines finds patterns and drafts opportunities. A person reviews the evidence and decides what should move forward.

How do I get the app or a new version?

For a first install, book a call — we share the current signed build after that conversation.

Book a call

After you are installed, use the in-app updater. If a newer build is required, Update installs through the signed updater. Product changes are listed on the changelog.

How do I delete everything on this computer?

Settings → Storage → reset local data, or uninstall and remove ~/.jarbas (the Pacific Machines data folder). Shared team reports and synced Brain documents already in the organization workspace are not removed by a local reset.

Who do I contact?

Product and privacy: hans@pacificmachines.ai. Sales and onboarding: book a call.

Support and updates

This guide is the public user documentation for the Pacific Machines desktop app. It is updated when product behavior that users rely on changes.

Related public pages:

• Product — what Pacific Machines is
• Changelog — release notes, newest first
• Status — public application status and incident communications
• Security — security whitepaper and how to report a vulnerability
• Privacy Policy — how information is collected, stored, and shared
• Terms of Service — terms that govern use of the app and this website

Contact: hans@pacificmachines.ai