ShogunAI

Product features

One context layer from memory to action

ShogunAI connects three parts of knowledge work that are usually separated: remembering what happened, retrieving the right context, and completing the next step.

01 · AI memory

It learns your work by watching you work

ShogunAI reads the text macOS already exposes to screen readers, keeps it encrypted on your Mac with its source and time, and indexes it on-device.

Learn more
02 · Contextual recall

Ask about your week without pasting the backstory

One question across mail, chat, docs and calendar, answered from your own history with the source and time attached to every hit.

Learn more
03 · Execution layer

Answering is the easy half

Drafts, updates and actions inside the tools you already use, under three named levels of autonomy — and a stop before anything reaches another person.

Learn more
ShogunAI — Full view

This is the app’s own interface, running here with sample data. Move through the panes.

Open in a new tab

In detail

What each part actually does.

The capability, and the rule that binds it. Nothing here is on a roadmap.

Capture

Text through the accessibility layer
The focused window’s app, title, and visible text, read the way a screen reader reads it.
RuleExcluded apps, private windows and password fields are never read. Pause stops it entirely.
Visual recall
A compressed screenshot, only where text capture returns nothing at all.
RuleOff until you turn it on. Encrypted on the Mac, deleted automatically at the retention age you pick.
Meeting audio
Streamed to a speech service for live transcription while the meeting runs, and never used to train anyone’s models.
RuleNever written to disk or a temp file. Only the transcript persists, and the traceability log records the egress.

Recall

One question across every tool
Mail, chat, docs and calendar answered from a single search.
RuleEvery hit carries its source and time, so an answer can be checked rather than trusted.
On-device embeddings
Indexing runs on your Mac with a bundled multilingual model.
RuleA heavy day costs nothing extra to index, and recall works with no network.
Tiered memory
The last day in memory, the last month in full detail, everything older compressed.
RuleThis morning answers at conversation speed without carrying a quarter of raw text.

Execution

Press Option
The continuation is written at your caret, in the app you are typing in.
RuleA local write. Nothing is sent — you send it.
Three levels of autonomy
Reversible work runs, drafts wait for you, sends stop for approval.
RuleAnything that reaches another person is level three. No prompt moves that line.
Traceability
What ran, on what evidence, and what left the device.
RuleThird-party hops are marked as such. The log never leaves the Mac.

Meetings

Detection and transcript
The meeting is noticed and transcribed while it happens.
RuleYou can exclude an app, a single meeting, or turn detection off entirely.
Minutes
Summary, decisions, and the commitments it heard — with next actions as proposals.
RuleA proposed action is never executed. One tap files it into your work state.
Two-way translation
Set the language each side speaks; each hears the other in their own.
RuleWhat persists afterwards is text, in both languages.

Credentials

Your existing plan
Run on the assistant subscription you already pay for.
RuleInside that plan’s own limits. Opt-in, and never by reading another app’s credentials.
Or your own key
Bring an API key instead and set your own ceiling.
RuleKeys live in the system keychain and nowhere else.
Switching providers
Change model provider whenever you want.
RuleYour memory is not tied to a model. You lose no history.
ShogunAI Command Layer

Every model you use starts from zero. Give it your week.

Connect once, keep one private memory on your Mac, and let ShogunAI finish the work from there.

Joining the early-access waitlist is completely free.

macOS · 7-day full trial · your memory stays on your device.